Code Book, part 1a (Core back-end programming)

« Return to Code Book table of contents

Back-end

Bootstrapping

Bootstrap flow diagram (a diagram showing the bootstrap process expressed a little differently)

(Click to enlarge)

The basic interaction chain for a request is the same as pretty much any PHP system:

Web browser > HTTP request (URL, GET/POST) > Web server > Web server configuration > PHP > Composr > HTTP response > Web browser

One complicating factor is the URLs use. By-default, Composr uses standard PHP URLs, e.g.:
site/index.php?page=downloads&type=view&id=3
for viewing a download.

However, we do support a number of URL Schemes, so potentially it could be something like:
site/downloads/view/3.htm
or:
site/pg/downloads/view/3
or even just:
site/downloads/view/3

For incoming requests we handle this all at the "Web server configuration" point. The vast majority of Composr websites are powered by Apache, so in this case it involves the rewrite rules .htaccess file (Apache's mod_rewrite module).

From Composr's point of view, the requests always look like standard PHP URLs, with GET parameters available from $_GET (although we use our get_param_string functions to retrieve these, for security reasons).

An Composr page view comes into an index.php file within a zone directory. In the above example it is the site zone, so site/index.php.

The index.php file contains standard initialisation code. It:

1. sets up the basic Composr file paths
2. sets the directory path to the base directory (so everything is harmonised across different Composr requests)
3. initialises the minimal bootstrap code sources/bootstrap.php
4. uses bootstrap to pass off control to sources/global.php (and injecting a PHP strict_types declaration if running in development mode)

global.php defines the very core of the Composr framework, in particular code needed to load up library files.
global.php calls up global2.php, as the first library file called using require_code (i.e. global2.php supports overrides to its code and has a standard initialisation function, global.php and bootstrap.php do not).

Next init__global2() runs, as this is the standard initialisation function for global2.php. This code loads up most Composr subsystems (such as the database connection). Lots of code files get loaded up (making APIs available), with lots of initialisation functions.

One of the code files is site.php. This code file checks zone and page permissions for the user, and will also be used to generate the main page view.

When init__global2() is finished, control pops back into the original index.php file, which calls do_site(). This proceeds to work out what page code to load (module, Comcode page, etc), and make Composr compose the page (putting it into the GLOBAL_HTML_WRAP.tpl template etc), and finally output it. To learn more about how Composr composes pages from its various elements, and how it decides what kinds of pages to load, see our Composr site structure.

We should also note our URL convention. Essentially we have up to 3 standard URL components: page, type, id. page identifies the Composr page to load (which can be any kind of page, it's found via a search process). type and id are specific to a particular module, but are a strong convention, and have special representation in the URL Schemes described above.

Philosophical points

Composr is written to the 'front controller' design pattern. The 'entry-point' scripts (front controllers) are what take in initial requests. In particular the index script takes in requests for any page within a zone. This is opposed to a system where each individual page is initialised from its own PHP file. This pattern is mainly there so that we can support non-PHP pages, such as Comcode Pages.

Our use of standard page/type/id parameters represents a design philosophy called "convention over configuration". Many PHP frameworks have custom designed routing flows, but we prefer to design a convention, then make our framework automatically carry that forward. It makes things simpler for everybody if basic and reasonable decisions are built into the system. This said, there is no reason you can't have your own routing schemes for your custom entry-points.

We try and stick fairly close to PHPs own architecture rather than forcing an entirely different request interaction pattern. So, we use the same system of reading GET/POST data from global state. There are cases where modularity dictates having simulated request states within the system, and we have special code for making that possible -- but we treat it as an exception rather than forcing all request interaction to go through a complex system that contradicts PHP's own design.

Configurable bootstrapping

Composr initialisation can be controlled by a few global variables in the entry-point script. They all default to false.

• $MICRO_BOOTUP (true will tell Composr to not use a user login, skip the ability for sophisticated interfaces, and skip permissions)
• $MICRO_AJAX_BOOTUP (true will skip a lot of stuff loading, intended for AJAX scripts that run a lot; this is intended to reduce server load)
• $KNOWN_UTF8 (true will tell Composr the request character set is definitely in Unicode, as the request came from JavaScript)
• $FORCE_INVISIBLE_GUEST (true will tell Composr to not use a member login)
• $CSRF_TOKENS (true will tell Composr to check for a CSRF-token if reading any POST-data; so only use this if CSRF-tokens are being set in calls to the script!)
• $STATIC_CACHE_ENABLED (true will tell Composr it can check the static cache for possible static-cached versions of the output)
• $IN_SELF_ROUTING_SCRIPT (true will tell Composr that standard Composr URLs can be routed through this script)

We do use global variables in Composr, even though many frown on them. Composr is optimised to the bone in terms of performance and code size. We have a rule that if a global variable is used in more than a few places that variable must be documented clearly within the code.

Addons

Composr is split into addons. Some addons are core, some are optional. Optional addons may be removed, and the removal results in file-deletion. Composr has inbuilt addon import and export support, that allows a user to use and distribute packaged addons for Composr ("non-bundled addons").

Example:

• Bundled core addon -- core_rich_media (the implementation of Comcode, basically)
• Bundled non-core addon -- downloads (file publishing / document management)
• Non-bundled addon -- hybridauth (Login using a Facebook account, etc)

Addons contain module-pages (and blocks), and it is these modules (and blocks) that define things like database structure. Actually addons may define this too directly in their registry file, but it is rare for us to do it this way. Modules (and blocks) are independently versioned, but the versions of individual modules (and blocks) are hidden from user -- the module versions exist so that modules have a way of tracking when they need to upgrade themselves (a mismatch between the version on-disk, and the installed version in the database will prompt a module to upgrade itself).

Composr knows an addon is installed if there is an addon_registry hook present. File lists for addons are defined in these hooks. Templates may work against the 'ADDON_INSTALLED' symbol, to guard dependencies; likewise, code may use the addon_installed function. In some cases, modularisation is enforced using hooks rather than extra logic; this is done in the case where there is a clear concept that can be created, which might be useful for other people to add into in the future (e.g. the side_stats blocks allows different addons to plug in hooks so their stats can be displayed). Third-party addons do not have so much luxury over adding extra code into the main Composr code, so must rely on either existing hooks, or code overriding.

Addons may define dependencies.

In the case where we have an addons [addon A] hook for another addon [addon B] (e.g. the galleries hook for the newsletter addon), we place the hook in addon A. This is because the hook would only run if addon B is there, so placing it in addon A is safe -- but if it was placed in addon B and addon A wasn't installed, it would fail unless we wrote extra code. Composr will auto create directories that do not exist -- directories aren't a part of addons, only individual files.

Config options that have dependencies between two modules (e.g. the points_ADD_DOWNLOAD option -- which needs points and downloads) should have a default value evaluation of null if the foreign dependency is not met. So for this example, the option is part of the downloads addon, and the default is null if the points addon is not installed.

User management of addons

If a user is installing an addon (from Admin Zone > Structure > Addons), it will install any modules/blocks within it, if said module/block was not already installed.

If a user is uninstalling an addon, it will uninstall any modules/blocks within it.

A user should upgrade an addon by installing it on top of itself. i.e. don't uninstall the old version, but install the new version on top. Composr detects not to do a fresh install by checking if it is already installed.

Two addons altering the same file (special case)

If you have two addons that are competing for overrides on the same file, there are two approaches to solve this:

1. you'd write an override in such a way that both addons would package that same override, and you'd use if (addon_installed('...')) to dynamically adjust how it runs depending on which of those are installed.
2. you would improve the core code, e.g. with additional hook support, so that both addons could hook in (required GitLab access).

Modules

What is a module?

In Composr terminology 'modules' are dynamic 'pages' .

Example: index.php?page=<module_name>

Each module has a number of 'screens', identified by a URL parameter named type.

Example: index.php?page=moduleName&type=<screen_name>

Generally features relating to a single system (e.g. 'downloads' system, 'galleries' system, 'awards' system, …) in Composr are contained in a single module, with the exception that content-management and administrative features are each put into separate modules (e.g. cms_downloads and 'admin_downloads') . Content-management (CMS) and Adminisative (admin) modules are made separately so they may be placed in separate zones. Typically a website might be configured to allow certain members to manage content via the CMS zone, but only staff may access the admin modules (via the Admin Zone).

Module files are either full-modules or mini-modules. All standard Composr modules are full-modules. Mini-modules are designed to simplify development or integration on individual websites, where full Composr coding quality is not important.
Henceforth any use of the word 'module' in this tutorial will be talking about a 'full-module'.

Like any Composr page, a Composr module resides in a zone. Most modules reside in the site zone.
Content-management modules are all prefixed cms_ and reside in the cms zone. Administrative modules are all prefixed admin_ and reside in the adminzone zone. The other zones are only used in very special cases.

Recall that Composr zones are sub-directories of your site, which operate with different settings. By default, Composr contains a number of zones:

• Welcome (/)
• Admin Zone (/adminzone) -- Where Composr is configured
• Site (/site) -- Where the majority of the Composr modules are, by default
• CMS (/cms) -- Where the content is managed
• Forum (/forum) -- Only for installations using Conversr

A Composr site webmaster can add as many additional zones as they wish. Additionally, addons may come with zones. For example, buildr (an online dungeon game system) comes with a buildr zone.

How do you create a new mini-module?

The Composr mini-module feature lets you create new Composr pages very easily. You don't need to code to any particular structure or API, just write plain PHP and output as normal from it. You have full access to Composr's APIs for when you need to interface with the rest of Composr.

Mini-modules allow:

• PHP programmers with no experience with Composr to hit the ground running
• experienced Composr developers to develop simple pages without any coding overhead
• easier porting of third-party PHP scripts into Composr (you may need to change some links around, and remove HTML header tags -- but it's a lot easier than doing a rewrite)

We will present 3 examples. To try each out, simply save the code into a site/pages/minimodules_custom/example_page.php file, then call it up via http://yourbaseurl/site/index.php?page=example_page.
As you can see, site/pages/minimodules_custom/example_page.php corresponds to a page called example_page, in the site zone.

As mini-modules are just a kind of Composr page, you can control access to them using normal Composr page permissions (i.e. set from Admin Zone > Security > Permissions tree editor).

Example 1

The standard introductory example, Hello World.

Example 2

Now let's do some simple Composr API calls.

Example 3

Need to output a simple spreadsheet? You are allowed to set headers and exit(); within your page, so that Composr doesn't continue doing anything more after your spreadsheet has output and your code has run.


Note that Composr has proper spreadsheet APIs, this is just a simple example that does not use them.

How do you create a new full-module?

A module is a PHP file that defines one class. For a page in the site zone named example, the file would be site/pages/modules/example.php (or site/pages/modules_custom/example.php if this module is non-official, or a modified version of an official module). Our module example would contain a class named Module_example. Usually this class does not inherit, although some modules do inherit from a class named Standard_crud_module, which is used for acceleration/standardising development of add/edit/delete user interfaces.

The info function

The module class must include a function named info, which returns a map of fields that describe the module:

• author -- this is a string indicating who authored the module. The author does not need to be defined anywhere else -- this field is only intended for human-reading.
• organisation -- like author, but this is the organisation that the author is working on behalf of. For modules developed and maintained by the Core Development Team, it is 'Composr'.
• hacked_by -- this should be left to null, unless the module is an edited version of someone else's module, in which case it would be the name of the person altering the module (the new author)
• hack_version -- this should be left to null, unless the module is an edited version of someone else's module, in which case it should start as the integer '1' and then be incremented for each new version of the module (see explanation of version)
• version -- this is the version number for the module. The version can start at any integer, but '1' is usual. This number should be incremented whenever a compatibility-breaking change is implemented.
• update_require_upgrade -- this should be either true or false (it defaults to false if it is left out). If it is true then the install function will be called when Composr detects that the module's version has increased since it was originally installed. It should only be set to true if the install function is written to be able to perform upgrades (this requires extra effort).
• locked -- this prevents administrators from uninstalling this module. This should be set to false unless the module is a core module that should never be uninstalled.

The run function

The module class must include a function named run. This function is called when the module is loaded up as a page. The function should never use echo or print commands, but rather, it returns page output in Tempcode format using the do_template function. Tempcode is essentially a tree structure of composed template output, and will be properly explained later in the tutorial.

The run function typically is written to do 4 things:

1. Loads-up/records dependencies that all screens in the module use, using the require_code/require_lang/require_css/require_javascript API calls.
2. Gets the URL parameter named type into the variable $type, usually giving a default type of browse for when no type parameter was specified. Remember that type indicates which 'screen' the module will be outputting.
3. Delegates to another function depending on $type.
4. Outputs the result of the function that was delegated to, or new Tempcode() if no delegate was found for the given type.

For example,

The get_entry_points function

This function returns a mapping that identifies all the supported 'types' (screens) that may be launched directly via URL (i.e. all the ones that don't rely on form submissions to have been sent to them). The mapping maps from the type to a language string codename that provides a human-readable label to describe the screen.
It is used in the Sitemap, which in turn is used by things like the menu editor (to help Composr website administrators find links to add to their menus).

For example,
returns a mapping for two entry-points: the browse screen (which almost any module will have), and the tree_view screen. ROOT and TREE are language strings.

The install and uninstall functions

Modules are typically responsible for the installation, upgrading, and uninstallation of the database (and some filesystem) parts of the system that they provide screens for.

For example, the downloads module sets up:

1. database tables that relate to downloads
2. database indexes that relate to downloads

and removes:

1. database tables that relate to downloads (and associated indexes)
2. entries in shared database tables that relate to downloads (access permissions and trackbacks)
3. uploaded files that relate to downloads
4. stored values that relate to downloads

Composr modules check if installation/upgrade is needed when accessed.
If a user is installing an addon, it will install any modules/blocks within it, if said module/block was not already installed.
If a user is uninstalling an addon, it will uninstall any modules/blocks within it.

If you are supplying an updated module to a user, outside of the formal addon process:

• If the database structure has changed, code the module with correct versioning to update the database structure (versioning is described further down)
• Make sure the user does not uninstall the old module, because we do not wish to lose the data
• Get the user to simply replace the old module file with your new one; just accessing it will then cause it to prompt to upgrade

The install function

All database access is Composr is performed through the database objects. For almost all situations, the $GLOBALS['SITE_DB'] object will be the one to be use.

Database tables are created using the create_table function of a database object. This function defines the schema of the table. This function takes two parameters:

1. The name of the table
2. A map between the field names, and Composr field-type-identifiers

Composr defines the following field types:

• AUTO, an auto incrementing unique key. This allows you to have a key field for the table and not have to worry about generating the keys yourself -- when you insert your data (query_insert function) for a new row, just leave out the key, and the key will automatically be created for you, and returned by query_insert.
• AUTO_LINK, a foreign key link referencing an AUTO field. It does not specify what table this field is in, but the code itself will know how to use it. It's usually obvious from the name chosen for the field (e.g. a field named category_id in the news table obviously is referencing the key of the news_categories table).
• INTEGER, an integer
• UINTEGER, an integer, greater than zero
• SHORT_INTEGER, a integer
• REAL, a float
• BINARY, 0 or 1 (i.e. a representation for a Boolean)
• MEMBER, a link to a member
• GROUP, a link to group
• TIME, a date and time (output of the PHP time() function)
• LONG_TRANS, a long piece of text stored in the language/comcode translation table
• SHORT_TRANS, a short piece of text stored in the language/comcode translation table (255 length maximum)
• SHORT_TEXT, a short non-translatable piece of text (255 length maximum)
• LONG_TEXT, a long non-translatable piece of text
• MINIID_TEXT, a very short piece of text (about 50 length maximum)
• ID_TEXT, a short piece of text (about 100 length maximum)
• IP, an IP address in string form
• LANGUAGE_NAME, a language identifier (e.g. EN)
• EMAIL, an e-mail address
• URLPATH, a URL or file path
• MD5, a MD5 hash, stored in base64encoded form (the output of the md5() function is as such)

If a field type has "*" in front of the name, then it will form the primary key. All tables must have a primary key on at least one field.

The field types (apart from the AUTO and various string types) may have a "?" put in front of them (E.g. "?INTEGER"). This indicates that the value may be null.

Note that Composr also comes with a database integrity scan (for MySQL). In order for this to work efficiently and accurately, some guidelines on field names must be followed depending on the field type used. This allows the integrity scan to correctly match database field types to Composr field types. Namely, the name of the field should contain the following phrase depending on the field type:

• All types except INTEGER and UINTEGER: Should not start with 'count_', end with '_count', or contain '_count_'.
• AUTO: Should always be named 'id'.
• AUTO_LINK: Should end with '_id' and not contain any of the phrases used for MEMBER or GROUP types.
• INTEGER, UINTEGER, SHORT_INTEGER: Should not contain 'author', '_id', or any phrases used for the MEMBER, GROUP, or TIME types.
• MEMBER: Should contain 'member', 'user', 'submitter', or 'owner' in the name, but should not contain 'author' ('author', or rather 'author_id', is an AUTO_LINK to an author by the addon of the same name).
• GROUP: Should contain 'group' in the name (but not 'grouping' as this is used by the forums).
• TIME: Should contain 'date', 'time', or 'until', but should not contain 'timeout' ('timeout' is an INTEGER or UINTEGER).
• LONG_TRANS, SHORT_TRANS: Should not contain any phrases used by the TIME type.
• SHORT_TEXT: Should not contain any phrases used by the URLPATH type.
• MINIID_TEXT: Should not contain any phrases used by the IP type.
• IP: Should contain 'ip_address' (and not simply 'ip').
• URLPATH: Should contain 'url'. Note that fields that support page-links should not be type URLPATH; use SHORT_TEXT instead and use 'link' in the name.

Composr minimises its use of database features, in order to increase portability. We intentionally do not use the following database features:

• stored procedures
• transactions
• default field values
• functions
• foreign key constraints, and automatic tidy up

On a practical level, these things aren't really necessary so long as the PHP code takes up the responsibility for providing the same kind of abilities instead. E.g. the model code for an addon would define default field values in how it pre-populates form fields for a new entry.

When tables are created, a special meta table is updated. This meta table stores the database schema in a database independent manner that allows the backup/restore system to work. Because of this, you should never manually change the database structure outside the context of the Composr database functions.

You might wonder why we don't just read the database schemas directly via abstraction code in the database drivers, and avoid the need for the special meta table. It would certainly make it easier to manage the database if this were the case. The reason is that our typing system (with the field types explained above) encodes semantic information beyond what simple SQL data types can encode, and this is of wider use within Composr. For example, the broken-URL scanner looks at all field values of type 'URLPATH'.

Database indexes are created using the create_index function of a database object. This function takes three parameters:

1. The name of the table to create the index on
2. A unique name for the index
3. A list of fields to include in the index

(Indexes provide important speed improvements if you routinely look-up data from tables using fields other than the primary key)

One important thing to understand at this point is that Composr supports multi-language content at its core. This means that any resources that are defined need to be able to define their human-readable strings via the translate table rather than directly. As of Composr 10, by default multi-language content is not enabled, but the code needs to support it anyway. This is the purpose behind the transtext and transline configuration option, as well as the 'SHORT_TRANS'/'LONG_TRANS'/'SHORT_TRANS__COMCODE'/'LONG_TRANS__COMCODE' field-type-identifiers.

If you store data in tables using 'SHORT/LONG_TRANS[__COMCODE]' you retrieve that data via calling the get_translated_text function upon the database field value.

There is one other reason the translate table is used, and that is so that content may be stored efficiently in 'Comcode'. Comcode is our dynamic markup language, a superset of XHTML which defines additional dynamic features. The translate table allows Comcode to be stored along with the parsed version of it, which is in Tempcode format. To get back the Tempcode, which can then be mixed in with templates, use the get_translated_tempcode function.

If multi-language content is not enabled all the same APIs are used, but the translate table is bypassed. String data (and where applicable, parsed Comcode) is saved directly into the normal database tables.

When developing a module you might find it easiest to set up the database tables by hand, and then code in the installer afterwards. Otherwise it can be time-consuming going backwards and forwards reinstalling modules every time a small change is required.

Be aware that the get_option function always returns a string regardless of the type of the config option. All config options are stored as strings, and the type only specifies the input mechanism.

Upgrading via the install function

The code in the install function needs to analyse the parameters passed to see if an upgrade is being performed (and from what version) or if a new install is being performed. The code is then structured to act accordingly.
For a new module there is no need to consider how upgrades would happen.

Your code will follow a pattern like:

The uninstall function

The uninstall function is usually very simple, with calls to the drop_table_if_exists function of a database object. Uninstall function takes no parameters and doesn't return a value.

A note about how methods are called

The following object methods are not actually called as methods, but actually as functions (using eval):

• info
• get_entry_points
• install
• uninstall

This means that you cannot use $this inside the code.

Architecturally this is not ideal, but for practical reasons it was necessary because of the PHP memory limit. Upon installation, or module querying (e.g. to build the site map), instantiating all objects fully would use a lot more memory than the PHP memory limit would allow.

Screen conventions

Modules functionality is subdivided by 'screen'. 'Screens' are segregated by their unique expected output -- individual error messages don't run from separate screen functions, because error messages aren't expected in advance, they're incidental.

As an example, if a module was designed to add a resource, it would need at least two screens:

1. The interface to allow the user to add the resource (a form screen). By convention this would be identified by type=add.
2. The actualiser to actually add the resource, and the screen to say that the resource had been added. By convention this would be identified by type=_add.

The function names of screens do not need to mirror the names of the type parameter, but usually it is clearer this way.

All modules must also define the default screen, which is almost always identified by type=browse or by there being no type parameter given. The role of the default screen depends on the module, but usually it either acts to display a category browser or a menu. Usually all other screens in a module will be reachable by navigating starting from the default screen of that module.

Config options

Configuration options are created using the config hooks. The hooks contain a get_details() function that defines a few elements:

1. A language string codename that identifies a human-readable name for the configuration option.
2. The codename for the configuration option.
3. A data type for the configuration option (this defines how the option is edited -- all options are stored and returned as strings). Valid data types are: integer, tick, line, float, transline, transtext, colour, forum, forum_grouping, usergroup, comcodeline, comcodetext, date, datetime.
4. Some PHP code that returns (in string format) what the default/initial value for the configuration option should be. Often it is just a static value, but sometimes it is calculated in an intelligent way. Alternatively, return boolean false if the config option is disabled for some reason, such as a missing dependency (disabled config options will not show in the configuration module).
5. A language string codename that identifies a human-readable name for the configuration category which the option belongs to. Many of these are 'FEATURE', which is the conventional set for all configuration options relating to optional Composr systems. Configuration sets do not need to be individually defined -- all sets that are referenced will be automatically listed and indexed.
6. A language string codename that identifies a human-readable name for the configuration group which the option belongs to. This often indicates the name of system that the addon is providing screens for.

Templates and Tempcode

Composr is basically a programming language within a programming language. The outside programming language is PHP, and the inside programming language is Tempcode. Tempcode is two things:

1. Composr's templating language/syntax
2. The name for instances of compiled Composr templates. These instances are instances of the 'Tempcode' object.

Tempcode (the language) is very rich and dynamic. It is designed so that templates can be infinitely complex but so that they can be cached so that the data that goes into them does not need to be recalculated each time. This caching usually only applies to blocks -- modules generally do not employ Tempcode caching. For example, the main_multi_content block might have been templated to look different for people in a certain usergroup, yet we also don't want to have to regenerate this block each time it is viewed. Tempcode is powerful enough to be able to represent arbitrary differences like this in a way that survives caching.

Due to this requirement, Composr must not deal in string output, it has to deal in Tempcode output. If we flatten Tempcode down to a string we would lose the scope for dynamic calculation within it. All through the system Tempcode is composed together, rather than strings.

A typical scenario for a module screen is as follows:

themes/default/templates/SUBCATEGORY.tpl:

themes/default/templates/CATEGORY_PAGE.tpl:

Tempcode (template) types

Be aware that Tempcode only supports these data types as template parameters:

• more Tempcode
• strings
• booleans (which are converted to strings as '1' or '0', which is what Tempcode uses for its boolean logic)
• arrays (but only for use by special directives -- referencing an array as a normal parameter will only return its size)

It intentionally does not support integers or floats, so these must be converted to strings manually when passed through, using strval (machine-read integers), integer_format (human-read integers), float_to_raw_string (machine-read floats), or float_format (human-read floats).

GUIDs

There is a design issue that comes to us when we design template structure… do we re-use templates so that editing is easier, load is decreased, and consistency is raised; or do we allow maximum customisation of different aspects of the system by not re-using templates?

We stick to a middle ground between these extremes, and re-use templates when the usage pattern is almost identical. For example, almost all of Composr uses the same form field templates, but Wiki+ posts use different templates to forum posts. However, there are still places where we re-use templates in situations that sites may wish to develop in separate directions.

Each do_template call in Composr, which loads up a template, may pass in a parameter, _GUID. The GUIDs may then be used by themers with tempcode directives to control what the template outputs.

Don't bother manually putting in GUIDs. We have a script that auto-generates them and inserts them across the code base. We run this script before major releases.

Making a simple two-screen full-module

We've discussed how to make a module with very basic screens and templates. Following is a more detailed example on how to:

1. create templates
2. link screens using page-links (build_url).
3. use Tempcode
4. read variables from GET/POST

Setting up our basic screens

To keep this code short I have intentionally missed out some functions and PHP-doc comments that would usually be defined, and not bothered using language strings. These should be included/used for real code, but aren't needed for the code to run.

Creating our templates

EXAMPLE_SCREEN_A.tpl…


EXAMPLE_SCREEN_B.tpl…

Blocks

Blocks in Composr are written almost identically to modules. They may have install/uninstall/info methods, which work in the same way, and a run method. The similarity is not surprising given blocks function almost the same as modules -- the only difference being that blocks are embedded within wider page layout, while modules defining whole screens (excluding panels etc).

The main five implementation differences and points of note are:

1. The run method takes a parameter, which is a map (array) of the parameters given to the block. You cannot assume any parameter is passed, so for each you must allow a default which you can hard-code into your block code.
2. You should define a BLOCK_blockname_DESCRIPTION and BLOCK_blockname_USE language string for the block, to explain what it does and what it is for.
3. The info method defines a parameters entry in the array returned. This is simply a list of the parameters the block can take. It is used by the block construction tool. For each parameter listed there should be a language string BLOCK_blockname_PARAM_parametername. Look at how existing block parameter language strings are done to see how to lay it out -- the format is parsed by the block construction tool so it is quite strict. In particular reference the default value in your string.
4. You can allow the block to cache by defining a caching_environment method. This returns an array which has a ttl (the number of minutes the cache lasts) and a cache_on which is a string containing PHP code that defines the signature (array) for a cached instance of the block. The PHP code can make use of $map, and must define a different array for each possible way the block could turn out (i.e. the array is a signature that distinguishes that particular appearance, based on the block's parameters and optionally other things like the viewing user's usergroup membership).

Blocks should be documented, and this is done by creating specially named language strings. These strings are utilised by the Block construction assistant to allow the webmaster to add in the block in a user friendly fashion. Imagine a block called example with parameters a and b. You would define language strings as follows:

As you can see, the parameter strings also define the default parameter values. This has to be done in a way that is structured/worded exactly like the above. It only affects the Block construction assistant, it doesn't affect the behaviour of the block when no parameter is passed (you should make that behaviour match what you define here though).

If you need to define a list then you have to do that in a very strict way, like:
BLOCK_example_PARAM_c=Foo. Value must be either 'x' or 'y' or 'z'. Default: 'x'.

Full reference for language strings

For blocks:

Language string	Purpose	Default
BLOCK_TRANS_NAME_*	Name of block	Automatic from block codename
BLOCK_*_DESCRIPTION	Description of block	Blank
BLOCK_*_USE	Purpose of block	Blank
BLOCK_*_PARAM_*_TITLE	Parameter title	Automatic from parameter codename
BLOCK_*_PARAM_*	Parameter description	Blank

For Comcode tags:

Language string	Purpose	Default
N/A	Name of tag	Automatic from tag codename
COMCODE_TAG_*_DESCRIPTION	Description of tag	Blank
N/A	Purpose of tag	N/A
COMCODE_TAG_*_PARAM_*_TITLE	Parameter title	Automatic from parameter codename
COMCODE_TAG_*_PARAM_*	Parameter description	Blank
COMCODE_TAG_*_EMBED_TITLE	Embed value title	"Tag Contents"
COMCODE_TAG_*_EMBED	Embed value description	Blank

(as you can see, there are no language strings for defining tag names or uses)

For both blocks and Comcode tags you can use any of the following language construction components…

Magic string within parameter description	Purpose	Applies for blocks	Applies for Comcode tags
" (an ID number, or name)"	Just a bit of text that will be stripped, because an input widget will make that obvious.	Yes	No
"Advanced feature. "	Put parameter under an "Advanced" tray.	Yes	Yes
"Whether (...)"	Used to indicate a checkbox should be used for input.	Yes	No
"hook-type: <hook-type>"	Used to indicate a drop-down for selection of a hook from that hook type.	Yes	No
"Default: 'some value'."	Used to specify the parameter default.	Yes	Yes
"Must be either ... or ... or ..."	Used to create an ad hoc list. Can be as many terms as desired.	Yes	No (but you can do a simple pipe-list instead)
"The number of "	Used to indicate a numeric input should be used.	Yes	No
" Supports Comcode."	Used to indicate that Comcode is supported within the given value.	No	Yes

Various of the above cases only work for blocks because Comcode tags tend to have dedicated interfaces built rather than interface hinting.
There are many cases for both Comcode tags and blocks where the code picks up on particular tag/block/parameter names and provides very-customised input widgets (or full forms).

Hooks

This section is an explanation of hooks in Composr, using search as an example of how they are used.

The search module is a general purpose search tool. It provides a UI and it shows results. It needs to be able to find different kinds of results.

When something in Composr needs to do something but in lots of different kinds of way, or a place needs to be defined for different things of the same kind to happen, hooks are used.

Hooks are modular. They are plugins. They prevent us having to hard-code too many different things in an ugly bloated way that would be impossible to extend or strip down.

Hooks that go together do the same kind of thing. In the case of search, each search hook searches a different content type -- but they all search.

All hooks that go together (e.g. all search hooks) have the same class interface and are stored in the same directory (*1). The code that uses the hook instantiates them and loads method(s) with parameter(s). The method(s) and parameter(s) used is essentially a design contract between hook implementations and the code written to call the hooks. They all fit a pattern. What pattern to use is defined by whatever calls the hooks -- in our case, the search module (*2); the actual hooks have to fit that pattern. Most hooks have an info method and a run method, but that is just a convention.

An example of some code that uses hooks is:

This code finds all modules/search hooks, then loops over the list of hook names (these are stored in the keys returned from the find_all_hooks function). In each loop iteration it:

1. loads up the code file for the hook
2. instantiates the object in the hook using Composr's object_factory function (which it can do because it assumes a class naming convention for all of these kind of hooks)
3. checks to see if the object instantiated okay, and if it didn't, continue to the next loop iteration. The true parameter to object_factory will prevent Composr automatically failing if it can not create an object and instead make it return null. We do this in case users accidentally upload some other kind of PHP file to the hook directory: if we didn't check for the error here, it'd have caused a stack trace as it wouldn't contain the class we needed.
4. calls the info function on the hook, loading its results in $info. Again, this is an assumption via a convention for these kinds of hooks, that they will have an info function that behaves in a certain way. From this particular example (search hook) our info function will return us an array structure of details about what the hook can do, and later the code will determine whether to put an 'advanced' search link in for the hook according to whether it supports advanced searches or not (I haven't shown that code here for conciseness).

(*1) Not quite. Hook files of the same kind might be in sources/hooks/<something>/<somethingelse> and also sources_custom/hooks/<something>/<somethingelse>.
This is because you place custom files in sources_custom so that they are easily identified as custom. It is very common in the case of hooks, if you're adding a new hook to Composr, because they are a key tool in the customisation/extension of Composr.

(*2) The Conversr members module does member searching, but does not use the search hooks. This module's member search is separate and hard-coded. There is still a search hook for member searching for the search module to use though.

You can generally tell what uses the hook by its file path. For example hooks in sources/hooks/modules/search are for the search module.

Another example: member profiles

Composr needs to be able to provide a spot for addons to put links on somebody's profile screen (e.g. maybe an eCommerce addon needs to put a link to their transaction history onto their profiles). So there is a kind of hook in Composr to achieve this. The contract the hook meets is to take a member ID and return a list of extra links to put in their profile.

The code that uses these hooks is as follows:

In this case you can see the code assumes the hooks have a ->run method that takes a member ID as a parameter, and returns an array. Again this is the 'contract' these hooks have to meet.

Let's look at one of our hook implementations to see what code it has in it. We will look at the calendar members hook (sources/hooks/members/calendar.php):

Notice how this class is named Hook_members_calendar. The calendar bit comes from the filename, the Hook_ bit is a standard contention for any hook, and the members_ bit is something all the members hooks are using as a convention and really is just there to stop namespace conflicts (e.g. if we also have a calendar hook for search we wouldn't want these two classes in Composr to have the name, so we try to make our sets of hooks use their own unique naming convention to keep them apart).

As discussed previously, these hooks have a convention of having a run method that takes a member ID as a parameter, and thus the hook implementation above has this. The code inside the method does 3 things:

1. It returns no-results if the calendar addon is not installed. We don't strictly need to do this, as this file won't be here normally if the calendar addon is not installed (the hook file is listed as one of the files in the calendar addon so would be uninstalled along with other calendar files). The reason we do it is in case the file was accidentally restored after the calendar addon was removed (there are many ways someone might accidentally do that).
2. It checks a couple of permissions. One is that the accessing user has permission to assume_any_member -- if they don't, no-results are returned, so the link won't show (this particular hook's link is for the purposes of admins, to allow them to access other member's calendars via their member profile, so should only show to admins really). It also checks the accessing user has access to the calendar module in general.
3. It then returns an array structure that matches the convention for these hooks. That is:
4. "List of tuples for results. Each tuple is: type,title,url" (checking the PHP-doc API signature is always handy if you want to know how a hook should behave). In this case the link is a 'views' link, is given a link caption of whatever the CALENDAR language string contains, and has given the build_url-defined URL. The members module (this is where the hooks are called from) knows what to do with this tuple of data, we just had to make sure we return what was expected.

So now that I have explained how hook's have contracts, how they are called, and gone through an example hook implementation, you should be able to write your own. To write a new hook simply:

1. take an existing hook file, copy it to a new filename (e.g. example.php instead of calendar.php)
2. change the class name in your file to reflect the new filename stub (e.g. example instead of calendar)
3. pull out the old code inside the hook's methods and write your own code to do whatever your new hook needs to do (so long as it fits inside the usage pattern/contract of the particular kind of hook you are working with)

Another example: symbols

Composr supports 'symbols' in templates, that do special things. An example of a symbol is our MEMBER_PROFILE_LINK symbol, which makes it easy to link to a member's profile screen from a template (often templates are passed member IDs as parameters, so it allows them to work with their parameters in an effective way). But Composr needs to be extendable. Addons must be able to add their own symbols (e.g. maybe an eCommerce addon would need to define a 'BALANCE' symbol to show a member's balance in any arbitrary template), so there are symbol hooks.

For performance Composr doesn't use any symbol hooks for any default functionality, but they are available for new addons.

Content types

In Composr content types are implemented via a lot of different files. The best way to find out how to add a content type is to search for all the files named after an existing content type. 'polls' is a good example as this is a fairly simple Composr content type, and these are the files for it (it is possible this is outdated):

• sources/hooks/systems/addon_registry/polls.php -- This is an addon-registry hook, and essentially just lists the files associated with the poll system/addon. You only need to worry about writing addon-registry hooks if you are writing code that is going to be bundled with the main Composr distribution. These hooks exist so that the addon manager can package/delete all files relating to polls.
• lang/EN/polls.ini -- This is the file defining the language strings for the content type. Most of the strings are just ones specific to the code for polls, but some will be used as standard by the CRUD module: ADD_POLL, EDIT_POLL, DELETE_POLL. For any poll strings to be used in code/templates the require_lang('polls'); command would need to be run in the PHP code first (often in a module's run function).
• site/pages/modules/polls.php -- This is the main site module (a kind of page) that provides the user-focused poll functionality. It provides screens for viewing polls, etc.
• sources/blocks/main_poll.php -- This is a block for showing the current poll. Webmasters may place it on their Comcode pages (such as their front page) via this Comcode:
  • [block]main_poll[/block]
• sources/hooks/blocks/main_staff_checklist/polls.php -- This is a staff checklist hook that is used to show whether the poll needs setting, as a checklist item on the Admin Zone dashboard. Most Content-Types don't provide this hook.
• sources/hooks/modules/admin_import_types/polls.php -- This is an importer hook, it's not very interesting and you are very unlikely to need to write one.
• sources/hooks/modules/admin_setupwizard/polls.php -- This is a Setup Wizard hook, and this particular hook allows the Setup Wizard to place the poll block automatically on the website (Setup Wizard hooks can do various things, this is just one example of what a Setup Wizard hook can do). You are very unlikely to need to write one of these hooks.
• sources/hooks/modules/search/polls.php -- This is a search hook, to allow polls to be searchable from the Composr search module.
• sources/hooks/systems/content_meta_aware/poll.php -- This hook is used to provide the search-engine-friendly URLs for polls. It specifies the details Composr needs to be able to automatically generate URL keywords and tie them into ID numbers. In the future it is likely Composr will use this kind of hook for more things, and can generally be considered as a way for a content type to declare/define some of its properties.
• sources/hooks/systems/page_groupings/polls.php -- This hook places poll icons in the Admin Zone/CMS Zone do-next menus.
• sources/hooks/systems/preview/poll.php -- This hook allows polls to have user-friendly previews from the add/edit form. It provides a method to scan to see if you are trying to preview an poll, and another method to generate a preview from the data passed to the preview script via the POST environment.
• sources/hooks/systems/rss/polls.php -- This hook provides an RSS feed for polls.
• sources/hooks/systems/trackback/polls.php -- This hook determines whether a specific poll has trackbacks enabled (this is used by the trackback-reception script, to check to see whether it has access to save the trackback).
• sources/polls.php -- This file defines the API functions relating to polls. For any poll API functions to be used in code the require_code('polls'); command would need to be run in the code first (often in a module's run function).
• sources/polls2.php -- This file defines rarer API functions relating to polls, mostly write-functions.
• themes/default/css/polls.css -- This file defines the CSS for polls. For any poll CSS styles to be used in a template the require_css('polls'); command would need to be run at some point in the PHP code that loaded up the templates (often in a module's run function).
• themes/default/images/icons/menu/social/polls.svg -- This is an icon for the menus.
• themes/default/templates_cached/EN/polls.css -- This is a cached CSS file, and is what the HTML will actually reference. It is generated when Composr automatically parses the Tempcode in the main polls.css file.
• cms/pages/modules/cms_polls.php -- This is the CMS module, that adds/edits/deletes polls. It's an CRUD module, and thus is structured in a standard way. From an MVC perspective it's the controller that glues the polls API (the model) and the forms interface (providing the view), defining form fields and reading/passing them for saving.
• (various templates) -- There are various templates relating to polls also. These are used by the block and by the screens in the site module and are specific to the content type.

Installation code

In Composr we need to program modules so that they auto-create database structure and database contents. Even if working direct for a single client, it is best that modules be constructed like this as it eases development (it's easy to set up different development sites for example).

If creating, for example, some kind of property website, it would be best to pre-populate location tables in the installation code of the new key module that uses them. The default catalogues module does a similar thing, with its creation of some default catalogues.

It is natural that you won't get your database code right first time. When developing you probably do not want to have to manually alter the database and code separately each time a non-minor change is made, although of course you can do. Instead you can reinstall the module from the Admin Zone. Go to Admin Zone > Structure > Addons, then you'll find a link for module management under there. If you find you can't uninstall/reinstall/install your module, you coded it as 'locked' (in the info method) and you probably shouldn't have.

If you find stack traces are generated by module management then some of your code is probably faulty. It might not be the same code as you expect (module management probes various different files), but look closely and you will probably find the error.

When developing you'll probably want to start your development site afresh a few times to test your code in different configurations, or to wipe test data. If you develop on a server that is not reachable from the public web, you can leave the installer in place so long as you have a blank file named install_ok in the base directory. This is better than having to move or rename the install.php file because if you do that you risk messing up its versioning, especially if you work out of a source code control system.

Testing

Composr has a multi-faceted testing strategy as there is no perfect solution for testing (we have thought very long and hard to find one!). Our overall strategy involves different approaches, largely integrated into our automated testing (testing_platform addon):

1. Code Quality Checker, a lint (henceforth 'CQC'). This is a standalone tool, but also integrated into our automated testing. See _tests/codechecker/readme.txt in the testing_platform addon for documentation.
2. Automated test framework, with a large number of tests written for it. The framework can incorporate  regression testing, as long as we write regression tests whenever we find bugs. It can incorporate UI testing in a sense as we can program a web bot to do things plus dump all the XHTML). The framework is documented in the description for the testing_platform addon.
3. Web standards Validation (partly built into CQC, partly built into Composr -- and also implemented as an automated test).
4. Manual functional test set. The main test set is no longer maintained due to the enormous amount of time required to manually go through the tests, while things likely to break are picked up pretty well by the automated testing. However, we also maintain a smaller list of important tests (https://composr.app/tracker/view.php?id=3383).
5. Custom version of PHP, with type strictness, and XSS detection, among other things.
6. Manual testing at core-team meet-ups, especially Usability testing / Blackbox testing.
7. Public beta testing and bug reporting.
8. Composr development mode, which applies lots of tricksy things, checking various things are done at run-time, and turning features on and off randomly, and doing some performance testing.
9. Other things we use: MySQL strict mode, Strict JavaScript errors.
10. Composr error e-mails (telemetry).

Software for editing code

For coding you'll want to use either an IDE or a text editor. IDEs are heavier but have more features.

We love:

• CodeLobster
• PhpStorm
• NetBeans

We have found the best text editors for Composr have been Notepad++ (Windows), TextMate (Mac), and 'Geany' (Linux).

The above are just suggestions. Use whatever you like (within reason -- not Windows Notepad).

Emulating TextMate

Composr has support for the TextMate txmt URL handler, for opening files from your local server referenced by Composr in a text editor. For example, when a stack trace hits, each trace line would have a link to open it in TextMate.

To enable this support you need to run this command in Commandr:

While this feature is fairly unique to TextMate, you can configure other text editors to work with it too, on different operating systems. We won't be documenting every text editor and operating system combination, but we have documented Geany on Gnome (Linux) below. For Windows there is a registry editing mechanism to register protocol handlers. Some browsers such as Firefox don't actually need operating system integration.

Geany on Gnome (Linux)

1. Create /usr/share/applications/txmt-uri.desktop (sudo nano /usr/share/applications/txmt-uri.desktop):
   

   Code (INI)

   [Desktop Entry]
   Name=Geany
   GenericName=Geany
   Comment=Open txmt links in Geany
   TryExec=open_geany
   Exec=open_geany %u
   Terminal=false
   Type=Application
   MimeType=x-scheme-handler/txmt
   NoDisplay=true
    
2. Run sudo update-desktop-database
3. Create /usr/local/bin/open_geany (sudo nano /usr/local/bin/open_geany):
   

   Code (Bash)

   #!/bin/bash
   FILE=$1
   FILE=${FILE/txmt\:\/\/open\?url\=file\:\/\//}
   LINE=$(echo $FILE | grep -o "\&line=[0-9]\+")
   LINE=$(echo $LINE | grep -o "[0-9]\+")
   FILE=$(echo $FILE | grep -o "\(.\+\)\&")
   FILE=$(echo $FILE | cut -d'&' -f1)
   geany +$LINE $FILE
    
4. Run sudo chmod a+x /usr/local/bin/open_geany
5. Relaunch your web browser
6. Go to wherever you are seeing txmt links, and click one (if you want to force it, edit the execute_temp.php to have a fatal_exit('!'); call in the place you put custom code)
7. If your browser doesn't integrate with Gnome automatically it should ask you what program to open the link with, manually select /usr/local/bin/open_geany

The .desktop file registers the protocol handler with Gnome. Browsers such as Firefox and Chrome should recognise whatever registration scheme your desktop environment is using, otherwise the final step will achieve the same end. The open_geany script remaps TextMate launch syntax to Geany launch syntax.

Javadoc-like commenting

All functions should be documented in the Javadoc-esque syntax used for other functions. Javadoc is a simple way of describing a function via a specially formatted comment placed just before the function starts. The syntax is parsed by Composr's own special functions so it isn't quite PHPDoc/Javadoc syntax but it is very similar.

With PHPDoc and the function header-line itself, every function has the following:

• A description
• A list of all parameters
  • The code-name of the parameter
  • The type of the parameter (including whether false [~type] or null [?type] values are accepted)
  • A description of the parameter
  • Whether the parameter is optional
• The return type (if any), and a description of it

The syntax is as follows:

As well as providing documentation, the API comments define function type signatures. This is used by our Code Quality Checker to help determine bugs. The types are based off the types used to define Composr database schema, and can be any of the following:

• AUTO_LINK (a link to a auto-increment table row)
• SHORT_INTEGER (an integer, 0-127)
• REAL (a float)
• BINARY (0 or 1)
• MEMBER (a member ID)
• GROUP (a usergroup ID)
• TIME (an integer timestamp)
• LONG_TEXT (a large block of db-targeted text)
• SHORT_TEXT (a block of text, up to 255 characters)
• ID_TEXT (text for a string identifier, up to 50 characters)
• IP (an IP address in string form)
• LANGUAGE_NAME (a two character language code)
• URLPATH (a URL)
• PATH (an absolute file path)
• MD5 (a string representation of an MD5 hash)
• EMAIL (an e-mail address)
• string
• integer
• array
• boolean
• float
• tempcode
• object
• resource
• mixed (this should rarely be used; but when is, means multiple types may come out of the parameter/function)

Question marks (?) may appear before types, meaning the value may be null. Null must always have a meaning add must be documented in a note in the parameter description… (null: <explanation>). Similarly tilde marks (~) may appear before types, meaning the value may be false, and this must also be commented like (false: <explanation>). Sometimes it is also useful to specify the meaning of the empty string like (blank: <explanation>).

Language files

Language files are .ini files under lang/<lang_name> or lang_custom/<lang_name>.

They follow the simple format:

We don't normally bother defining a descriptions section, but I included it here for completeness. Descriptions explain what a language string is for in cases where it is not completely clear or where misunderstandings could develop.

Error handling

Error-checking policy (automatic raising of errors)

There's no need in Composr code to listen for every possible error that PHP might give back. You can use some common sense, only detecting errors that have a cause, rather than trying to detect those that can't theoretically happen. Errors that can't theoretically happen, but happen anyway, will trigger Composr's automatic error mechanism and we can then deal with them as they come up.

For example:

• File permissions may always be wrong, so we should check return values / suppress and handle write-errors.
• We can assume things like critical Composr directories are present, such as sources
• We can't assume non-standard directories like 'lang/FR' are present, so we must either add some conditional logic, or cleanly handle the error event in an appropriate way (in this case perhaps explaining to the user that 'The French language pack is not installed').

Suppressing errors

Composr will react fatally on most errors unless they are suppressed. This includes PHP_NOTICEs. To suppress errors from a PHP command, put an @ symbol before the function call (although highly not recommended; you should handle errors responsibly instead). If the PHP command causing the error is buried inside a Composr API function and you want errors to be shown as 'attached' rather than fatal, you can call push_suppress_error_death(true) then pop_suppress_error_death() after.

Manually raising errors / showing messages

To raise an error you'll typically call one of these functions:

• trigger_error -- routed through PHP error handling (which itself is taken up by Composr, with logging and error notifications).
  • Error levels (Notice, Warn, Error) are PHP-specific.
• relay_error_notification -- just send an error notification
• error_log -- direct logging to the Composr error log (via the PHP error logging system)
• cms_error_log -- as above, but also includes an error notification
• fatal_exit -- some kind of particularly unexpected/egregious error, that needs to come with a stack trace and bail Composr out completely. Includes logging and error notifications.
• warn_exit -- A general error with a warning message to the user.
• inform_exit -- not actually an error, but exit with a message.
• attach_message -- an error is shown in the global layout, processing otherwise continues as normal. You can generate inform/notice/warn messages.
  • inform -- Used to inform the user of general / neutral information that does not typically require action. The message will appear in the theme's seed color.
  • notice -- Used to tell the user something that needs their attention but is itself not an error. The message will be seeded towards yellow.
  • warn -- Used to tell the user something went wrong (think warn_exit, but without the need to exit). The message will be seeded towards red.

Use trigger_error/fatal_exit if it is a totally unexpected but detectable error condition (i.e. not user error, not a common system error, i.e. very likely a Composr bug). As such it would be subject to logging and error notifications. As trigger_error goes through the Composr's handling of the PHP error system, the configuration options for how to handle PHP errors are respected. In many cases trigger_error will effectively call fatal_exit/attach_message based on these settings. The other functions are to be called directly in detectable situations that are typically user errors, where you want to specify the precise handling for them.

warn_exit and attach_message can be told to do logging and error notifications also via setting a special parameter. In this case though no error e-mail will send to the developers, as it's not a Composr bug.

Function	Error level	Handling type	Logging	Error notification	Developer error e-mail
trigger_error	Notice, Warn, Error	Depends on config	Yes	Yes	Yes
relay_error_notification	N/A	Skip	No	Yes	Yes
error_log	N/A	Skip	Yes	No	No
cms_error_log	N/A	Skip	Yes	Yes	No
fatal_exit	Error	Fatal	Yes	Yes	Yes
warn_exit	Warn	Fatal	Depends on parameters	Depends on parameters	No
inform_exit	Inform	Fatal	No	No	No
attach_message	Inform, Notice, or Warn	Attach	Depends on parameters	Depends on parameters	No

Getting error messages

Error handling in PHP is very inconsistent.

This section does not directly/specially relate to Composr, but we thought we would add it to help new PHP programmers. Generally we deal with what we need to, in the way we need to, within the Composr framework -- and you need to do whatever you need to do for PHP within your own code.

There are 3 main ways to get an error message in PHP, depending on what PHP function you are calling:

1. use a *_last_error() function
2. suppress the error with '@', check return values, then call cms_error_get_last()
3. catch exceptions (some PHP features only, typically newer ones)

Security

Escaping

A source of great confusion in programming is all the different kinds of 'escaping' that occur. Because the web is a great fusion of technologies, all kinds of things operate together with their own encodings, and when lumped together, it can be a mess unless you truly understand it all. And understanding escaping is the key to secure and robust coding.

Therefore I will summarise the kinds of escaping here…

JavaScript/PHP escaping

If you want to output a string in JavaScript/PHP that contains quotes, you need to prefix the quote with a '\' symbol. The same applies for '\' symbols themselves. The reason for this is because strings themselves are bounded by these quotes and hence to actually use these quotes within these quotes they need to be distinguished. This convention was taken from C and is used in many languages.
For example,
would output, Hello 'reader'.

SQL string escaping

SQL encloses string values for fields inside quotes, which presents the same issue PHP had. With Composr coding the db_escape_string function abstracts the process of SQL escaping. Any manually assembled queries should be assembled very carefully using db_escape_string (to escape strings) and intval (to enforce integers not to contain string data).

SQL escaping is key to PHP security, as it prevents data leaking out into the query itself, which could potentially allow users to rewrite the query to a malicious form.

HTML escaping

HTML uses the characters < > " ' and & with special meaning: they help define the HTML tags themselves. Thus if these symbols need to be referred to directly, things called 'entities' are employed instead. Entities are a special HTML syntax that encode characters which the standard character set, or the natural of HTML, would normally preclude from use.

Important entities are:

• &nbsp, to get hard-spaces (prevents wrapping or trimming)
• &gt;, replaces > ('greater than')
• &lt;, replaces < ('less than')
• &quot;, replaces "
• &#<ascii-code>;, an ASCII-code-defined character
• &copy;, an example of a symbol that wouldn't ordinarily be usable as it isn't in the character set (this is a copyright symbol)

Composr has the means to perform escaping of template parameters, so if you pass escape data to a Composr template, and program the template to escape it itself, you will see the entity codes instead of the entity replacements.

Escaping all data embedded in HTML is a key part of security.

URL parameter escaping

Parameters in URLs are separated by '&'. Keys and values and separated by '='. Also spaces are not allowed, among many other characters. If you pass any complex data that will go into a URL, use the urlencode function (encodeURIComponent in JavaScript) which will sort all this for you. The Composr build_url function handles it all in a much better way though, so you should actually almost always use this.

Vulnerability types

Security is treated very seriously by the developers. We have implemented a number of abstractions and policies to keep Composr as secure as possible and it is crucial that developers understand security issues and our policies.

Even admin-only sections should not have unintended exploitable behaviour. If a user is hosted in a restricted Composr-only environment, their software should not allow circumvention of current environment. Also, if a user exploits their way into somewhere they shouldn't be, it is best that somewhere shouldn't allow them to exploit themselves into somewhere even worse… for example, if Composr's Admin Zone gets hacked, a user should not be able to get full PHP code execution access on a hosting account.

There are a number of notorious ways PHP scripts such as Composr can be attacked, including:

• SQL injection
• Logic errors / assumptions
• XSS attack
• Header poisoning
• Code uploading
• File accessing
• Backup stealing
• Jumping-in
• Cross-Site-Request-Forgery
• Eval-poisoning
• preg_replace code insertion

We will briefly discuss many of these below…

SQL injection is prevented by the database abstraction layer. It is impossible to inject SQL through this layer as the SQL calls are made up by a formal conversion from PHP parameter syntax to SQL syntax by code that fully considers escaping strings and handling the-given data types accordingly. Any manually assembled queries should be assembled very carefully using the db_escape_string function as appropriate (to escape strings).

A good tip that isn't strictly related to security, but may be useful for general stability: if you are doing an update or delete, and you know you're only doing it to one record, put a limit of '1' on the query. This limits the damage that coding errors might cause, and it also limits the impact some possible security holes could have.

One common logic error that deserves to be mentioned is the false assumption that you can check for resource ownership by equating the logged in user against the member-ID of the resource. The reason for this is it doesn't cover the case of the guest user -- the guest user ID ('1') is shared between all guests. Therefore handle guests as a special case and never give them any access over guest-submitted resources.

Another common (and very remedial) logic error is assuming that a hacker won't try and bypass the UI to gain access by manipulating GET/POST variables. Security checks need implementing both in UI code and in actualiser code that carries out the actual tasks.

Another common logic error is passing in secondary ID values from a form, for edit/deletion in actualiser code, and only checking permissions on a primary ID. For example, if a record has some subrecords, and there are checkboxes to delete those subrecords available on the main record's edit screen, it is easy to forget that you need to run permissions checks on the subrecords as well as the main record. As an alternative to running extra permission checks, sometimes it is possible to construct queries on the subrecords that hard-codes the primary ID, such that the query will only execute if the subrecords referenced really do belong to the main record we checked permissions for; this is valid protection also, and usually involves less code.

Any special and obvious hack-attack situation that you can figure out (whether real or not) should be logged using the log_hack_attack_and_die function, so that site staff may detect when people are trying to hack or probe the site. While hackers with the source code may avoid the attack scenario you've predicted and logged, it is likely they will make mistakes and get logged anyway, which can be a huge benefit to site admins.

XSS is short-hand for "cross-site-scripting", and essentially poisons a user's HTML output. There are two common ways this can occur:

1. through URLs
2. through submitted data

The general premise is that through some kind of invalid input, JavaScript gets inserted into the HTML output. This kind of hack is indirect -- invalid output is generated via the vulnerability, then someone with wide website access is tricked, or naturally directed, to view that output.
The JavaScript in the output usually does something particularly malicious, like redirecting someone's password cookie to a hackers website -- allowing the hacker to steal their credentials.
Composr protects against this kind of attack in three major ways

• easy escaping syntax in templates
• parameter constraints (taking in integers instead of strings, for example)
• HTML filtering / Comcode
• our own custom version of PHP that we develop on, that can automatically detect these vulnerabilities in any code whenever that code is run (yes we can do his amazingly enough -- it is a very cool technology!)

Every value in a template that is not meant to contain HTML is marked as an escaped value ({VALUE*}). This means that 'html entities' are put in replacement of HTML control characters, making it impossible for such data to contain HTML tags.

Header poisoning is a technique whereby malicious header information is inserted to totally bypass HTML restrictions. Because the attack operates at the header level, it is important that headers with new-line symbols do not ever get inserted . An example attack would be to inject a complex header that causes (a) HTML output to begin early, and then (b) some special JavaScript to be injected (e.g. \r\n\r\n<script>alert('hack');</script>).

Code uploading is when malicious programs are placed on the server using Composr's attachment system. Such attacks are prevented by file type filtering, inside the upload abstraction code -- so always use the get_url function to handle uploads.

Also, never allow file paths to be given for editing or viewing in a URL without sanitising them. For example, the download system doesn't allow you to specify the local URL of a file that is not publicly downloadable.

If any kind of file path component comes in from a taintable source (e.g. user input) it creates a potential for a file accessing vulnerability. Make sure anything untrustable that goes into a file path is filtered via the filter_naughty or filter_naughty_harsh functions.

Pay attention to any kind of backups, make sure the paths are cryptographically secure so that third parties cannot download them.

Jumping-in is a term I just coined for this guide to refer to when a .php file is accessed by URL that was only designed to be included. Never put code outside functions in Composr, unless that script is intended to be called by URL. Composr has an init__<file> function naming convention to handle the situation where included files need to run some startup code -- the require_code function runs these functions automatically, simulating the effect of loose code.

CSRF (Cross-Site-Request-Forgery). It is easy to forget we encode actions in forms… potentially destructive actions which are called by the automated click of a button. The person clicking a button on a form has no idea where that form leads unless there is some text to say, and that text may not be accurate. A hacker could create a form on their own website that is encoded to submit malicious requests to a victim webmaster's own website. The hacker would then just need to trick the webmaster to fill that form in via some form of deception ("fill in this form on my website to win a free iPod"). Because the webmaster has access to their own site, any form they fill in also does by extension and thus the submission of the malicious form could cause real destruction before the webmaster realised what had happened.
The hacker in this situation has just performed a destructive action by proxy. Sometimes these hacks are elaborate in that XSS was used to make a user not even know they were clicking a link in the first place (and thus the hacker staying potentially anonymous).
We can't completely prevent this type of attack… a user might be tricked into filling in a form on someone's website that directs to a manipulative URL on their own. It can only be advised that users be careful what they click… i.e. they don't go to dubious websites or click links given in dubious circumstances, or click links with apparent malicious or strange intent.
It might be suggested that the REFERER be checked to make sure the REFERER for an admin action is always inside the users site, but the REFERER field is often changed by privacy aware users to be non-representative, and thus cannot be relied upon. Nevertheless, Composr does do this if the referer field is provided.
One action Composr does take, is to refuse to allow a user to click-through to an internal Admin Zone URL directly from an external site if they are not already logged in… it requires an explicit login for every new browser window. It might seem that this is an unnecessary hassle, but it truly is the only standard/environment-compliant way to prevent automated Admin Zone-actions. Users may disable this feature by turning it off when they edit a zone, and likewise turn it on for other zones.

Some nosy or malicious users may choose to try and poke around in upload, or page folders, by manually specifying URLs. They may attempt directory browsing, or to guess the URL to a file. Composr prevents directory browsing by placing index.html files in folders that shouldn't be browsable, thus nullifying the browse-ability with a blank output (one cannot assume the web server is configured to have directory browsing to disabled). In addition, .htaccess files are placed in folders where files should not be accessed by URL at all, but this only works on Apache servers. The upload system supports obfuscation of upload names in order to make URL guessing unrealistic.

It goes without saying, but I'll say it anyway, that any code that finds its way into an eval call should only come directly from hard-coded Composr code itself. Never do something stupid like:
eval('add_image('.get_param_string('image_id').');');

This code would allow any user to inject malicious PHP code into the image_id variable. Never use variables inside eval() unless you're positive that no user could have contaminated its value.

On a similar note, when using user-input with the PHP preg_replace function, always make sure that it is properly escaped, as the preg_replace function can actually do arbitrary code execution if input is not properly escaped.

HTML escaping in Composr

We need to escape any plain-text string that ends up encapsulated within HTML to avoid "XSS injection" vulnerabilities (and potentially corrupt output too, depending on what is inside the plain-text string).

There are 4 basic cases for escaping being done in Composr:

1. Generally escaping is performed inside templates (using the asterisk escaper, *) because generally we pass raw data to a template. This is because in principle a template should be able to do whatever it wants with the data. That may be showing it within the template output, that may be doing calculations/checks on the data, or any other undefined thing.
2. There are cases in the API of mixed-type parameters where escaping may or may not be done, depending on whether a string was passed (escaping is performed) or Tempcode was passed (no escaping is performed as it is assumed the output of a template and thus already in HTML format). This behaviour is clearly and consistently documented in the PHPDoc (provided in plain-text format or HTML Tempcode). This pattern happens when this is the behaviour the programmer would want in the vast majority of cases.
3. There are cases in the API where escaping may or may not be done for you depending on if you set a function parameter to do so, combined with the above (i.e. for string parameters only). For example, the results_entry function takes an array of $values but a single $auto_escape setting that will escape all string values given if enabled. This is documented in the PHPDoc as format depends on $auto_escape (string or Tempcode). This pattern happens when it is reasonably likely the programmer may want to manually escape some values (via escape_html) but not others depending on which of the values being passed are plain strings (i.e. need escaping) or (already) HTML rather than just based on string vs Tempcode. By convention $auto_escape parameters are never optional so that you need to actually consider the behaviour you want.
4. There are cases in the API where there is no escaping, leaving escaping up to the user of the API via manual use of escape_html. This is documented in the PHPDoc as provided in HTML format (string or Tempcode). This pattern happens when the programmer would want to be passing HTML in the vast majority of cases.

Complexities around language strings

Language strings may or may not be in HTML format. The do_lang function is usually used for when a language string is plain text, while do_lang_tempcode is usually used when a language string is HTML.

There may be situations when a template escapes a parameter (case 1), but for some rare situation we are passing in HTML to that parameter and want to avoid escaping. Any output of do_lang_tempcode is automatically immune from escaping for this reason. For example, imagine we have a template that is showing a value in a simple table row…
Normally we want BAR to be auto-escaped, as it is a simple value, hence why we used the asterisk escaper. But imagine we wanted to pass a language string for <em>N/A</em> as BAR. We would need that to not be escaped. Hence we have a magical exception to the asterisk escaper: do_lang_tempcode output is never escaped by it. This is handled 'magically' by the Tempcode system, which can track if something is a language string.

We also have a function protect_from_escaping that achieves the same results for cases where you need something to not be escaped which is not a language string, and under-the-hood it actually uses do_lang_tempcode with a loop-through language string to work.

This behaviour does create a need for a special consideration though: we need to ensure any plain-text parameter to do_lang_tempcode is escaped, otherwise they will flow through as unescaped to the output.

For API functions that take mixed parameters but assume plain-text, but we can conceive of a minority of use cases where HTML may be desired, this is explicitly documented in the PHPDoc as provided in plain-text format or as HTML via do_lang_tempcode/protect_from_escaping (string or Tempcode).

XSS injection detection in ocProducts PHP

ocProducts PHP can track if strings are escaped or not. Tempcode integrates with this and will complain if an unescaped string ends up in the output. This will help you avoid mistakes relating to the complexities described above.

Unfortunately this cannot pick up on the case of parameters for do_lang_tempcode not being escaped, for technical reasons.

Kid-gloves modes

It's a fact of life that novice coders don't know how to write secure code, and it's also a fact of life novice developers will be expected to write real-world production code. Composr contains many security features and abstractions to protect common types of vulnerability, however there are also additional protections included for novice developers.
The additional protections may slow performance slightly, or cause some slight weird processing issues in rare events, but generally won't cause any problems with simple code. The protections automatically enable for any block or module in a *_custom directory.
The protections may be turned off by a code call like:
This call serves philosophically as a self-certification, and practically turns off the kid-glove mode features related to what the developer is certifying against.
This way Composr has the best of both worlds:

• Experienced developers who write logical and correct code will not have to fight the system for rare corner case bugs that look like they could possibly be security-related.
• Newbie developers will be protected for the vast majority of cases.

Saying I_UNDERSTAND_SQL_INJECTION disables extra scanning of queries to ensure they are always made in the most secure way and that custom data within the queries is correctly escaped. The scanning slightly impacts performance.
Saying I_UNDERSTAND_XSS disables automatic escaping of template parameters and language string parameters (which can result in accidental double-escaping in output) and any input parameters that make it into the output of a mini-module/mini-block. It is smart enough to know what was and was not already escaped (due to internal tracking), so double-escaping is rare.
Saying I_UNDERSTAND_PATH_INJECTION disables hack-attack errors from being generated if any URL parameter has ".." or null bytes in it.

Think of these flags as a trade-off between safe code and correct functionality in all cases. Without the flags, code is safe but may not always run as expected -- with them, code is only safe if you know what you are doing, and will always run correctly.

Input sanitisation, the right way

Sanitising input is almost trivial in Composr. Just use the most appropriate reader function and Composr will then apply appropriate checking/filtering for you.
Reader functions include:

• get_param_string -- reading in a string from GET (i.e. the URL)
• get_param_integer -- reading in an integer from GET (i.e. the URL)
• post_param_string -- reading in a string from POST (i.e. a form)
• post_param_integer -- reading in an integer from POST (i.e. a form)

We also some separate security functions for preventing things like filesystem attacks, or basic conformance checking:

• filter_naughty -- make sure no file paths are in something that will be used as a path component
• filter_naughty_harsh -- make sure something is strictly alphanumeric
• is_alphanumeric -- an alphanumeric test that doesn't result in bail-out if failed
• is_valid_email_address -- an e-mail address sanitisation check (valid syntax, does not check if it is real, although is_mail_bounced can help find if an e-mail address has bounced in the past)

For database queries (i.e. to avoid SQL injection) use the simplified database APIs wherever appropriate, query_select for example.

To avoid XSS attacks ensure you use the correct logical escaping in the templates. For example, plain text used in an HTML context would be escaping like {EXAMPLE*} (i.e. with the asterisk).

CSRF attacks are already avoided via a form token system and referral checking system.

Stability

There are many important programming considerations that need to be applied in most projects, including Composr. This section discussions some of them:

• If you are using an editable field as a foreign key in some way (e.g. you have an editable codename field, and you reference that codename in another table to link things to it), then remember:
  • if the field is renamed then anything referencing it will also need to be updated to reference the new name. Composr can't do this automatically due to limitations in MySQL.
  • any key will obviously need to have its uniqueness maintained, so remember upon adding and renaming you must check that a new name will not cause a conflict
• Whenever you use a foreign key, you should make sure that anything dependent on it is fixed should the referenced row be deleted. This might involve deleting such dependent rows, or updating them to reference a different key. There is one exception in Composr: key integrity for member rows is not maintained for practical reasons. This means that you can never assume in Composr that a member ID stored somewhere (e.g. as a submitter) is still valid.
• Check for possible error values, especially when working with files. For example, if you're writing to a file, what if you run out of disk quota? What if an uploaded file is corrupt? Composr should handle such situations gracefully.

Type strictness

Type strictness comes naturally in most traditional programming languages, but languages like PHP and JavaScript are 'weak typed' and automatically perform conversions on the fly. For simple systems this is good as it makes things quicker and easier, but for complex systems like Composr it poses a problem.

As a strict rule we always use strict typing. The custom version of PHP the developers use enforces this.

There are a number of advantages to strict typing:

• Special tools can be written to scan code for errors, and these work much better if they are allowed to infer relationships via correct type information. Our own Code Quality Checker does this. For example, if we get parameter order wrong in a function call (e.g. to strpos) the checker can identify that as the type signature of the function call does not match the type signature of the strpos function.
• Similar to the above, our own type-strict version of PHP will catch more mistakes with strict-typing. Sometimes the mistakes will be a result of strict-typing not being followed, but usually it will be something like a parameter being missed-out, or the wrong variable being used.
• It allows us better security. For example, if a string is being put into an expression that forms the first parameter of the preg_replace function where we think we are passing a number, it could be a huge security hole (as it could be used to do arbitrary code execution). The programmer might have thought they were using an integer but because they used get_param_string instead of get_param_integer to read in a value they in fact did not. Because Composr enforces strict-typing these kind of mistakes get picked up on much more often.
• It discourages sloppy thinking.
• We have very close attention to detail. We don't like to throw numbers into templates and just have them output directly -- we want them to have commas for thousands, for example (depending on locale). Forcing explicit type conversions makes us consider these kinds of issues where otherwise they would easily be forgotten.
• Many databases are type-strict, but MySQL is not. Thus for developers programming on MySQL it is easy to just completely forget about these issues because code will always work on their development environment.
• It makes it easier to port Composr, should we ever need to (who knows what the future has in store, PHP might not survive for ever, or it might be surpassed by another language).

So, how does strict-typing work? In PHP each variable does have a type, it's just PHP does not normally stop you from, for example, using a string in place of an integer, or vice-versa (it converts a string holding an integer to a proper integer and vice-versa). With strict-typing we force ourselves to change variables to the right type using PHP's functions like strval.

Consider the following:

This is a perfect example of how nasty PHP can be in the wrong hands, as "Value is false" is written out four times (each time).

For Composr we would expect the following:

• Do not treat values as a boolean unless they are a boolean. Usually any single variable is going to have a certain type (in this example this is not the case, but usually it is and should be). To check if something is null use $something === null. To check something is blank use $something == ''. To check if something is false use !$something. To check if something is zero use $something == 0. To check if something is defined in an array use array_key_exists. To check if something is both defined in an array and non-null use isset. To check something is all of non-null, defined in an array (if applicable), and non-blank, use empty.
• Don't do $something == null. As mentioned above, use $something === null (is of the null value and type).
• If you have a variable that really is mixed type (like $x in this example) and you need to do a boolean check just for the case it is a boolean variable, do either is_bool($x) && (!$x) or $x === false.
• Don't use Tempcode objects as strings. These are not strings, and shouldn't be used as strings unless you do something like $string = $tempcode->evaluate(); to get the string equivalent first. It does actually work, due to __toString being implemented on the Tempcode object, but it makes the code harder to analyse.

Example strict-typing mistakes

If you have a list (an array with numeric keys, such as one returned from the query function), don't access the keys as if they are strings. E.g. $rows['1'] should be $rows[1].

Avoid this kind of thing, where one variable might be of more than one type…

If a numeric variable has no value, set it to null, like…
and then use === null checks/is_integer functions as required. We do not consider null a type like the others, due to how databases interpret it. We consider it more like a special marker meaning 'no value'.

If a variable really must have multiple types, do…
as this will flag the situation to our tools.

Another example:
List arrays have numeric indices, so it is wrong to reference them as strings like the above code does.

Another example:
This is a good example of why type strictness is important to us. This will create an array of strings, and the DB will be modified using strings -- however the database uses integers. So this code won't work on MySQL strict mode, or other database software. This is better…

Feedback

Have a suggestion? Report an issue on the tracker.