Php extensions

Editor's Notes

• #2: I mentor PHP developers into becoming C developers – because other PHP devs have done that for me If you want to move on and do more with extensions after this talk please talk to me – I also sit on freenode all day and answer questions and give feedback – always always want more code monkeys and fresh blood in any project and looking at extensions makes me sad – 99% suck
• #3: Give my short talk about why learning “down the stack” is so important and why more developers need to stop asking should I learn python or perl it’s better to learn in a domain entirely different than what you are currently doing if you do desktop learn web, if you learn web do desktop or mobile if you do PHP, don’t learn “across” (perl, python, ruby) learn DOWN java, C#, go, rust obj-c, C, c++ assembler the old gurus of C who built what we play with are dying off (do the math, how old is rasmus?) there won’t be hand holding, but a good mentor (for now me) will help you learn what you need to know
• #4: PHP 7 will be breaking all the rules here, changing things in a fundamental manner, just something to keep in mind We’re working on a project to make migrating to PHP7 well documented and easy when the time comes and also working on ways to improve extensions in general
• #5: This is the reason we have PHP extensions to hit the lower level parts of the code that are so important But the top reason is the #1 reason we write PHP extensions
• #6: This is usually the first thing people think of when they want to “get involved with PHP internals” But you will have a far easier time of things if you start with extensions
• #7: This is the proper path to get involved in PHP and the language - learning each piece of the stack step by step will make you not only a better program, and understand the language but it will make you a better part of the community
• #8: This is the #1 argument I hear – but really is not anything you need to worry about c98 people! you can use some c99 stuff but declarations MUST go at the top of blocks (that’s really good coding standards anyway) although this will be moot when Microsoft finally gets it’s CTP for 2015 just like php itself turn all your errors on when compiling (-wall is your friend) and try to code clean you don’t have to “know C” anymore then you have to “know PHP” to write a wordpress plugin if you have the very basics of what to do it’s not hard PHP takes care of a lot of the heavy lifting – from how to parse parameters coming in and how to shove data into someplace going out… to how to do fancy objects Unless you’re doing something REALLY evil (an opcode cache, changing the way the engine works, threading) most of this stuff has already been done for another extension, it’s just a matter of finding the code – compile and test btw, all the code generators out there currently suck – most don’t do test generation, doc generation, or use the proper apis
• #9: This is the book you want it’s affectionately called the C bible if you walk through this book and do the exercise you’ll get the basics of C
• #10: These are some pretty good free resources for learning C – walking through the two books and the online tutorial will get you a good portion of the way there remember programming is thinking as much as writing code as if you play piano or dance you must practice practice practice
• #11: So you’ll need to actually have an environment set up to compile PHP on I will say right now for your first foray into extensions you will be happiest on linux a VM is fine, in a debian/ubuntuish flavor because you’ll get a lot more help and support Also programming for windows and OSX is something you’ll need to learn before you finalize that extension but are much harder (yes, even OSX is harder – to do it right on there you use homebrew or something similar to install a linux stack – would be easier at first just to set up a linux vm)
• #12: This is the same and yet different on every system and we’re not going to go into it a whole lot – this should have been homework before you came ;) Basically you need a compiler – xcode on mac, vc(2008) express on windows, gcc something on linux – this is different for every system You also need the sdk and headers for your system – those will be put on with xcode, you’ll need the 6.1 windows sdk for windows Then you need autotools – bison, and re2c to build PHP On mac or linux you can use phpize to build extensions – You CAN do this on windows as well except that for some stupid reason the proper files are not shipped with windows binary builds (stupid) Then you need any depedencies (to build PHP as a base you need iconv, zlib, and libxml2 at a bare minimum) After you build PHP on windows you can use phpize (it will be generated for you) and makes life so much easier
• #13: make a note that windows does not need the ./ before configure and uses nmake instead of make (otherwise identical) Note that to use a specific PHP install use /full/path/to/phpize and –with-php-config=/usr/local/php5/bin/php-config Note this does shared extensions only – then again most of the extensions you’ll work on should probably be shared
• #14: So although I wont’ whine about it today – if you really want to do extensions properly you need to have your own php compile you need to not be afraid of gcc and clang you need to make sure you test and test and test
• #15: There, now we’ve switched to using a precompiled debug zts version of php 7.0 WOOHOO There aren’t (yet) other box versions/providers available – this is ongoing work (more on gophp7-ext later)
• #16: I’m putting parts of this code onto my github account – each section I talk about – scaffolding, adding a function, ext, will have a different branch – feel free to clone and play These aren’t necessarily complicated extensions but they do follow PHP CS, have all the copy and paste you need and do compile (I think)
• #17: So the part we’re all excited about – let’s write an extension! I’ll just get some code and then you copy and compile it and then…
• #18: If you’ve ever seen another PHP extensions talk they dive into how to use the zend engine and how things work and such PHP is to C as wordpress is to PHP – makes any idiot able to write a plugin for it ;) you don’t necessarily HAVE to know how the internals work – you just have to know the right calls to make! So although I might tell you about SOME of the internals how and why, for the most part this will be “how to do it” not “why you do it” Because PHP is a ball of rusty nails ;)
• #19: In addition this is the absolutely wrong way to approach writing PHP extensions! Actually it’s the wrong way to approach writing LOTS of things do you start out writing a website worrying about how to make a db connection before you’ve designed the database and created the schema? do you create the schema before you have wireframes? just because we’re making a product in a different language doesn’t mean the rules change
• #20: this is why you really need to think about what you’re doing – being able to use the extension is not really enough… just wrapping a C api is a good way to give yourselves headaches of a horrible nature Please be the solution, not the problem We need no more painful APIs in PHP
• #21: so these are the reasons you’d write a PHP extension as we talked about before – you want to use a C library, you want to make something process intensive faster, you want to hook into the engine itself there aren’t other reasons to do an extension wait? what is FFI?
• #22: So many languages support this idea of calling into usually C code Then they usually put a layer of regular code “on top” to make APIS non painful This can be useful it also tends to be slower
• #23: before you get into extension writing – if you just want an ffi wrapper and are just going to call the exact C calls from an existing library why go to the trouble of writing an extension? ffi is pretty great but a bit of a flakey extension yet, but it’s identical to python’s “ctypes” which is a stupid name, it’s really ffi I hear al lthe time about how “great” python is because of ctypes, frankly I beg to differ. Part of wrappign a C extension is translating the C calls into something far more “phpish”
• #24: oh ffi is cool and so needs a maintainer
• #25: so this is what we want to do – write an extension that wraps a good C uri parsing library
• #26: so you need to do your research right? there are multiple C parsing libraries, joyent has one and there are some nice c++ ones, but for simple well laid out C uriparser seems to win – also it’s really nice use your internet tools – when you started PHP I’m sure you had books and stuff on speed dial you will need to do the same for C
• #27: SO this is a VERY nice C library for parsing uris according to the spec – USEFUL AS ALL GETOUT cross platform, well maintained, easy to use – don’t let the sourceforge URL fool you, it’s in git 
• #28: C++ APIs or similar wrappers in python, perl, ruby, etc are a better place to look for api ideas – try a couple of ways of using the apis as well really think about what kinds of apis and code you’re looking for – so I was looking at wrapping a small library (uriparser.sourceforge.net) and one of the things I did first was sit down and write some rough almost pseudo-code of what the extension might look like
• #29: uriparser.sourceforge.net – a simple uri parser library that strictly conforms to rfc questions to ask yourself as you write these examples do I want namespaces? will this be an object oriented api or a procedural style api? do I want to do the extra work of a dual api? (such as date or mysqli?) these decisions are usually pretty easy to figure out once you actually try to write the kind of code you want to work
• #30: PHP is 90% copy and paste, the main copy and paste stuff is all scaffolding for an extension so we’ve figure out what we want to write and how we want to write it, now we need to have scaffolding in place for it talk about our 4 files, the scaffold example - show the files quickly to show what normally goes in a header, some PHP CS rules, et al
• #31: DO BOTH unless you’re absolutely certain that whatever you’re writing will only work on one system or the other m4 is autotools stuff, but the w32 are very easy to write as well, they’re just jscript (windows javascript dialect) with some special functions for things like checking headers and adding libs, etc talk about with vs. enable
• #32: this is the very basic items you can have in your struct to define what you module is and what it does STANDARD_MODULE_HEADER_EX allows you to add ini entries and depedencies to your module definition using STANDARD_MODULE_PROPERTIES_EX allows you to add globals what the globals are, a ginit and a gshutdown – plus a “post deactivate” method that basically never gets used ;)
• #33: Explain what these are to the C uniunitiated
• #35: Just some general rules for you to take a look at even if your extension is not going into PHP proper the more time you take to properly document your code the happier you will be
• #36: how to see if what you’ve been playing with will actually work This should load your boring extension without doing anything fancy, just allow it to show up in the modules list you can actually do a test case for this with extension_loaded or get_loaded_modules if you’re so inclined
• #37: everybody should ALWAYS test their extensions – test test test turn debugging on, turn memory leaks on - run your tests if they fail you’ll get a bunch of extra files from the tests system available that make things very easy to debug, a diff file, .out file, even an .sh file to run the test exactly like it was run the first time no .bat file yet… yes I have a patch for that
• #38: These are very basic functional style tests there are a bunch of different sections for each one – the important thing to remember is always include a skipif anything in “tests” will be run so it’s ok to put things in directories for order write tests for everything you can think of! this is my standard test to check my phpinfo (MINFO) function
• #39: Look at the code and look at the compile
• #40: NOTE: again there were many underlying changes made to php’s zvals for php 7 I’m not going to go into all the changes but I may mention a few we’re going to detour just a moment to learn something about PHP’s guts
• #41: But wait – PHP is sitting on C we are somehow translating C types to PHP types – how is this magic accomplished?
• #42: Our zval is a giant struct of doom for those of you who don’t know what a union is, A union, is a collection of variables of different types, just like a structure. However, with unions, you can only store information in one field at any one time. You can picture a union as like a chunk of memory that is used to store variables of different types. Once a new value is assigned to a field, the existing data is wiped over with the new data. This is what everything you do in C goes into!
• #43: PHP 7 has totally flipped around how a zval is structured
• #44: PHP 7 has totally flipped around how a zval is structured Simple values do not require allocation and don’t use refcounting. There is no more double refcounting. In the object case, only the refcount in the object is used now. Because the refcount is now stored in the value itself, the value can be shared independently of the zval structure. A string can be used both in a zval and a hashtable key. There is a lot less indirection, i.e. the number of pointers you need to follow to get to a value is lower. Discuss heap vs. stack memory here
• #45: So basically with an extension (especially one wrapping a library) what you’re doing is taking C variables and exposing them to php land by turning them into zvals so a “zval” in PHP has a union that can have the value of what is inside – something you’ll need to do when you create your extension is figure out how to “translate” that to a PHP type and how to handle it
• #46: In C, strings are simple NULL-terminated char arrays, like you know. However, when designing a scripting language such as PHP, the need to manage those strings arises. In management, we think about classical operations, such as concatenation, extension or truncation ; and eventually more advanced concepts, such as special allocation mechanisms, string interning or string compression. As length are typed on a size_t variable, they weight {platform size} bytes ! Whatever the platform. One of the PHP 5 problems is then solved : under a CPU64, string length will be 8 bytes (64 digits) for every platform (this is one of the definition of the C size_t type). Strings are refcounted ! In PHP 7, strings are refcounted (as well as many other primitive types). That means that interned strings are still relevant, but less : PHP layers can now pass strings from one to the other, as strings are refcounted, we are plainly sure that noone will accidentaly free the string as this latter is still used elsewhere (until doing an error on purpose).
• #47: p is for pointer (that’s good enough for me) these are helper macros to get the data OUT of that union of doom that you need, it’s also how you get zval types and other fun stuff note there are even MORE then this, but these are the ones you’ll use most often PHP is governed by macros and more macros, which is why it’s important to look at existing PHP extensions since 95% of the macros are NOT documented The PP macros have been ripped out of php7 because we no longer have pointer pointers all over the place And the macros without the pointer are used a LOT more than they used to be Booleans have internally changed – you should always check for IS_FALSE and IS_TRUE instead of considering as a Boolean value, and doing an lval on it will break
• #48: Now let’s make our extension like… actually do something
• #49: So we’re going to actually write a function for our uriparser library that does something extremely simple – we’re going to register a function to tell us the uriparser library version
• #50: Here show all the lines required to do one thing – link into the library mention how other options are thin on the ground, although cmake is gainin some major traction about 2008 there was an attempt to port php’s build system to cmake sadly cmake was not really ready for it it IS ready for it now (hhvm and others all use cmake) But again that’s a monstrous project and you’re fighting a battle gains? one build system for everything ,not maintaining our own configure system for windows, easier integration into ecosystems
• #51: you need to do three things to add a function to your extension define it’s arguments – this is for reflection define it’s body put it inside a function struct arginfo must have the begin and end arginfo then you add additional arguments as desired in between if you have optional arguments, use begin_arginfo_ex you can send by val or reference or do typehinting with arginfo as well
• #52: SO a lot of this is implementation detail issues so in PHP you use this macro – all it does is rename your function name to something that won’t collide evilly, and define a bunch of C parameters you’re getting passed in 
• #53: there are even more than this in php7 strings move to use the more correct The IS_LONG type now uses a zend_long value instead of an ordinary C long. The reason behind this is that on 64-bit Windows (LLP64) a long is only 32-bit wide, so PHP 5 ended up always using 32-bit numbers on Windows. PHP 7 will allow you to use 64-bit numbers if you’re on an 64-bit operating system, even if that operating system is Windows.
• #54: so you can also do “zend_parse_parameters_ex” the only thing you can do it with is QUIET there are a couple of reasons for doing this – mostly nesting and overloaded calls Note that variable args have dropped a pointer!
• #55: This is macro hell – PHP is macro hell every PHP function (or method) has a return value available to you – an automatically NULL zval that you need ot manipulate in order to give information back it’s called return_value you manipulate these return value zvals to get data back into PHP userland the one thing that is Retval_res used to be retval_resource Retval string no longer has the duplicate option
• #56: When you start to want to return complex data things get a little more complicated – this is where array and object init come in there are also ways to array_init with a specific length or object_init_ex which allows you to choose the type of object you’re creating there are a bunch more of these zend_API.h is your lifeblood for how to use these things, along with Note that strings used to have the option of saying “duplicate me” or not, now you don’t worry about it, because you’re using a zend_string with refcounts and such
• #57: Look at the code and look at the compile make clean, switch branch, You have to install uriparser by handls ls That means download and build – it has a pretty straightforward system You can –disable-doc and –disable- I forget (check in code) to ignore bad things
• #58: Now we move on to more complex parts of the code, methods that really DO something remember we should start by writing tests for it
• #59: Zend_object can be a variable length now! These can lead to far more complex code if you’re doing evil with objects We don’t really go over object handlers and other crazy stuff because that’s beyond the scope here of our learning, but something to look up
• #60: TO do a class – create and register it’s CE in minit put it in another file – will keep yourself sane instead of having 300 mile long files some people whine that this makes it harder to use static stuff, since each file is compiled into it’s own “unit” and then linked but modern compilers and linkers make the overhead on this moot – better to be organized!
• #61: so this is how you would do your minit function – this would go into php_uriparser.c where the other code displayed would be in uri.c – where all our code for our uri class is going it is generally a good idea to keep arginfo and methods to gether as well in the code, so when one changes you remember to change the other remember stuff must be declared in order to be used too! the namespace declaration and the PHP_MINIT declaration go into the header file I often do a private header file for things like that – or do an api file if I allow things in my extension to be used by say other extensions
• #62: So this has ONLY the basics – explain how the minit works then let’s talk lifecycle
• #63: Deep breath – let’s talk about those things we put in the module struct Each are run at a different point in time let’s also talk a wee bit about structure? sapis?
• #64: so this is the lifecycle of PHP each of these items has a different important role to play what you need to remember is that minit and mshutdown are called in the opposite order that they are loaded, but you can’t depend on other extensions being around unless you have registered deps (remember way back in the main struct when we talked about deps? this flow is also true for prefork php webservers But then we get into threads
• #65: When we add threading it gets a little more difficult, because minit/mshutdown are only called when apache is restarted there’s also ginit and gshutdown but they’re broken
• #66: There are multiple declarations to help with this – the ZEND_STRS is a helper macro – again macro hell
• #67: So this has ONLY the basics – explain how the minit works then let’s talk lifecycle
• #68: Class constants are another thing that’s pretty easy to do – global constants are evil, do NOT do
• #69: then we add some constants for fun
• #70: Now let’s make our extension like… actually do something
• #71: Methods have several components but look very close to functions that we did earlier. There is a getThis which is available inside that will give you the zval for the class notice how similar this is to functions!
• #72: go over the method stuff just a bit
• #73: getTHis is the way you get “this” – notice that no matter how you declare your properties in MINIT, you can “juggle” the vale around depending on what you set it to
• #74: go over the method stuff just a bit
• #75: Note – traits in extensions are basically broken 
• #76: these are some more advanced topics
• #77: A PHP extension's globals are more properly called the "extension state", since most modules must remember what they're doing between function calls. The "counter" extension is a perfect example of this need: The basic interface calls for a counter with a persistent value. A programmer new to Zend and PHP might do something like this in counter.c to store that value: This is changed significantly in the way this is stored – as we’re using thread local globals in our thread safe build
• #78: efree, emalloc, ecalloc – you can find good documentation on these manual/en/internals2.memory.management.php zend is taking care of all the memory stuff for you – allocating a larger chunk, cleaning up at the end of the request (but still, efree dammit) Often, memory needs to be allocated for longer than the duration of a single request. These types of allocations, called persistent allocations because they persist beyond the end of a request, could be performed using the traditional memory allocators because these do not add the additional per-request information used by ZendMM. Sometimes, however, it's not known until runtime whether a particular allocation will need to be persistent or not, so ZendMM exports a set of helper macros that act just like the other memory allocation functions, but have an additional parameter at the end to indicate persistence. If you genuinely want a persistent allocation, this parameter should be set to one, in which case the request will be passed through to the traditional malloc() family of allocators. If runtime logic has determined that this block does not need to be persistent however, this parameter may be set to zero, and the call will be channeled to the perrequest memory allocator functions.
• #79: you’ll do a lot of zend_fetch_object stuff to pick this stuff
• #80: Documentation is probably the most important part of the process, but I’ll still leave it for last If you’ve been writing examples and tests and protos and arginfo all along as I’ve whined about, you’re already 90 percent of the way there with docs, seriously the easiest way to do this would be to use a generator, however there aren’t any GOOD generators available right now so just go to svn and copy another extensions stuff – everything is created with PhD so you can test stuff for pecl you can put your code anywhere, for phpdoc you need to get your docs roughed out then ask them to be put in svn also mention readthedocs and the fact that pecl has licensing restrictions and we need a packagist
• #81: Go into here about how resources work (an integer pointer to a global table holding resources… or doing some other evil resource storage – basically you’re storing and throwing about pointers and they’re slow (DO NOT) ini entries are a horrible thing – the use cases are extremely small threading and such I can rant about for quite a bit and finally you can override parts of the engine from extensions – this is why xdebug and things like intercept work but they’re a bit beyond the scope of this talk
• #82: There is SOOO much more you can do from hooking objects to hooking the engine!
• #83: There is SOOO much more you can do from hooking objects to hooking the engine!