Everything
php
Learn Community Releases The PHP Project
Home » Project » Newsgroups » php.internals » ext/gd: support creation of animated GIFs »

ext/gd: support creation of animated GIFs

php.internals

Christoph Becker

9 years ago
Hi! A while ago I've grabbed up <https://bugs.php.net/32803> and submitted PR #2024[1]. This lay dormant until recently when Joe had a look at so many PRs (thanks!), so I'm bringing this issue to your attention. I presume nobody is opposed that ext/gd should support the creation of animated GIFs (they appear to be still in fashion again), but I'm not sure about the API. My first shot (as implemented in the PR) was to offer a rather minimal layer over gdImageGifAnimBeginCtx(), gdImageGifAnimAddCtx() and gdImageGifAnimEndCtx()[2] (you can see an usage example in the submitted PHPT[3]). Kalle raised some objections regarding the direct use of streams in the API, namely that imagegifanimbegin() expects an open stream to be passed as parameter, so perhaps it would be better to accept a stream URL instead and manage the stream behind the scenes. That appears to be much more solid (as the developer couldn't fiddle with the stream), but would require some encapsulation mechanism, either a resource, what would be in line with the other GD resource types[4], or an (opaque) object, what appears more suitable for PHP 7. Currently, I'd prefer an *opaque* object which would be created by imagegifanimbegin() (creating the respective stream internally), passed to imageanimadd() and be destroyed by imagegifanimend(). Before proceeding to the implementation, I'd like to hear your thoughts about that! [1] <https://github.com/php/php-src/pull/2024> [2] <https://libgd.github.io/manuals/2.2.3/files/gd_gif_out-c.html> [3] <https://github.com/php/php-src/pull/2024/files#diff-bbb4c7a8497032f98d062ac418d9fbbe> [4] <http://php.net/manual/en/image.resources.php>
-- Christoph M. Becker

Joe Watkins

9 years ago
Morning Christoph, We have to face facts: Resources are still a thing. While there is talk of switching resources to use opaque objects, I'm worried about introducing a mixture of objects and resources in random places throughout core exts, and having exts with both seems very strange to me. I know that this has already started to happen, but before it goes any further I would implore you to stick to resources for now. There needs to be an organised effort for some future version of PHP to remove the resource type and replace it with opaque (or not) objects. Cheers Joe On Sun, Jan 8, 2017 at 11:53 PM, Christoph M. Becker <cmbecker69@gmx.de> wrote:

Andrew Faulds

9 years ago
Hi Joe, Joe Watkins wrote:
> > We have to face facts: Resources are still a thing. > > While there is talk of switching resources to use opaque objects, I'm > worried about introducing a mixture of objects and resources in random > places throughout core exts, and having exts with both seems very strange > to me. > > I know that this has already started to happen, but before it goes any > further I would implore you to stick to resources for now.
The more things are made objects now, the less things need to be converted to be objects later. It might be “strange” to mix them, but would it actually cause confusion? So far as the user cares, resources and objects returned by Gd could just be integers. They're just opaque handles. Thanks.
-- Andrea Faulds https://ajf.me/

Christoph Becker

9 years ago
On 09.01.2017 at 21:21, Andrea Faulds wrote:
> Joe Watkins wrote: > >> We have to face facts: Resources are still a thing. >> >> While there is talk of switching resources to use opaque objects, I'm >> worried about introducing a mixture of objects and resources in random >> places throughout core exts, and having exts with both seems very strange >> to me.
ACK.
>> I know that this has already started to happen, but before it goes any >> further I would implore you to stick to resources for now. > > The more things are made objects now, the less things need to be > converted to be objects later.
ACK.
> It might be “strange” to mix them, but would it actually cause > confusion? So far as the user cares, resources and objects returned by > Gd could just be integers. They're just opaque handles.
I think that they *should* be opaque handles, but you never know how one's going to treat them. Presumably the most crucial point nowadays are type declarations; changing a resource to be an object could cause massive BC break (think of streams) – changing all resources to objects appears to be even impracticable for a major version. Therefore I think this has to be done in small steps ("eating the elephpant one bite at a time"), and so introducing new opaque handles as objects appears somewhat sensible even if that causes a mix of resources and objects in a single extension. Anyhow, I understand that the animated GIF support will need an RFC, and there should be a secondary voting option whether the API should use a resource or an opaque object.
-- Christoph M. Becker

Andrew Faulds

9 years ago
Hi, Christoph M. Becker wrote:
> Kalle raised some objections regarding the direct use of streams in the > API, namely that imagegifanimbegin() expects an open stream to be passed > as parameter, so perhaps it would be better to accept a stream URL > instead and manage the stream behind the scenes. That appears to be > much more solid (as the developer couldn't fiddle with the stream), but > would require some encapsulation mechanism, either a resource, what > would be in line with the other GD resource types[4], or an (opaque) > object, what appears more suitable for PHP 7. > > Currently, I'd prefer an *opaque* object which would be created by > imagegifanimbegin() (creating the respective stream internally), passed > to imageanimadd() and be destroyed by imagegifanimend().
Would this opaque object still allow you to use an arbitrary stream of your choice? Also, like with imagepng() etc., could you stream the output to uh… PHP's default output stream? (I'm not sure what it's called. The thing that `echo` goes to.) If it does both, I don't have much objection. Mind you… is it really necessary to hide the stream from the user? fopen() is not particularly hard to use, and there are (admittedly niche) cases where you might want to handle the stream yourself. Thanks.
-- Andrea Faulds https://ajf.me/

Christoph Becker

9 years ago
On 09.01.2017 at 21:19, Andrea Faulds wrote:
> Christoph M. Becker wrote: > >> Kalle raised some objections regarding the direct use of streams in the >> API, namely that imagegifanimbegin() expects an open stream to be passed >> as parameter, so perhaps it would be better to accept a stream URL >> instead and manage the stream behind the scenes. That appears to be >> much more solid (as the developer couldn't fiddle with the stream), but >> would require some encapsulation mechanism, either a resource, what >> would be in line with the other GD resource types[4], or an (opaque) >> object, what appears more suitable for PHP 7. >> >> Currently, I'd prefer an *opaque* object which would be created by >> imagegifanimbegin() (creating the respective stream internally), passed >> to imageanimadd() and be destroyed by imagegifanimend(). > > Would this opaque object still allow you to use an arbitrary stream of > your choice? > > Also, like with imagepng() etc., could you stream the output to uh… > PHP's default output stream? (I'm not sure what it's called. The thing > that `echo` goes to.)
It would be possible to allow either a stream URL or a stream to given, what would fit to image(png|jpeg|…)(). However, these are "stand-alone" functions which close the stream when they're finished, while the animated GIF writing functions would *allow* the stream to be manipulated by the userland developer, even though I can't see practical reason to do so. And, of course, we could also let the functions accept NULL to directly write to stdout. Passing in 'php://stdout' should have the same effect.
> If it does both, I don't have much objection. > > Mind you… is it really necessary to hide the stream from the user? > fopen() is not particularly hard to use, and there are (admittedly > niche) cases where you might want to handle the stream yourself.
Are there really such cases? Should we sacrifice the safety of internal stream handling to support some (hypothetical?) use cases?
-- Christoph M. Becker

Andrew Faulds

9 years ago
Hi, Christoph M. Becker wrote:
> On 09.01.2017 at 21:19, Andrea Faulds wrote: > >> Would this opaque object still allow you to use an arbitrary stream of >> your choice? >> >> Also, like with imagepng() etc., could you stream the output to uh… >> PHP's default output stream? (I'm not sure what it's called. The thing >> that `echo` goes to.) > > It would be possible to allow either a stream URL or a stream to given, > what would fit to image(png|jpeg|…)(). However, these are "stand-alone" > functions which close the stream when they're finished, while the > animated GIF writing functions would *allow* the stream to be > manipulated by the userland developer, even though I can't see practical > reason to do so. > > And, of course, we could also let the functions accept NULL to directly > write to stdout. Passing in 'php://stdout' should have the same effect.
Alright.
> >> If it does both, I don't have much objection. >> >> Mind you… is it really necessary to hide the stream from the user? >> fopen() is not particularly hard to use, and there are (admittedly >> niche) cases where you might want to handle the stream yourself. > > Are there really such cases? Should we sacrifice the safety of internal > stream handling to support some (hypothetical?) use cases? >
In fairness, I'm not sure. I know animated GIF streaming has been tried before, though I doubt if it's actually used at all these days. A more reasonable case might be embedding an animated GIF in other binary data. But that's a hypothetical. What is the actual “safety” gain of not using streams? That you don't need to use fopen() yourself? (PHP will close files for you.) I'm just not super convinced there's a need for adding this extra layer. Thanks for your response.
-- Andrea Faulds https://ajf.me/

Christoph Becker

9 years ago
Hi! On 10.01.2017 at 20:24, Andrea Faulds wrote:
> Christoph M. Becker wrote: > >> On 09.01.2017 at 21:19, Andrea Faulds wrote: >> >>> Would this opaque object still allow you to use an arbitrary stream of >>> your choice? >>> >>> Also, like with imagepng() etc., could you stream the output to uh… >>> PHP's default output stream? (I'm not sure what it's called. The thing >>> that `echo` goes to.) >> >> It would be possible to allow either a stream URL or a stream to given, >> what would fit to image(png|jpeg|…)(). However, these are "stand-alone" >> functions which close the stream when they're finished, while the >> animated GIF writing functions would *allow* the stream to be >> manipulated by the userland developer, even though I can't see practical >> reason to do so. >> >> And, of course, we could also let the functions accept NULL to directly >> write to stdout. Passing in 'php://stdout' should have the same effect. > > Alright. > >> >>> If it does both, I don't have much objection. >>> >>> Mind you… is it really necessary to hide the stream from the user? >>> fopen() is not particularly hard to use, and there are (admittedly >>> niche) cases where you might want to handle the stream yourself. >> >> Are there really such cases? Should we sacrifice the safety of internal >> stream handling to support some (hypothetical?) use cases? > > In fairness, I'm not sure. I know animated GIF streaming has been tried > before, though I doubt if it's actually used at all these days. A more > reasonable case might be embedding an animated GIF in other binary data. > But that's a hypothetical.
Okay.
> What is the actual “safety” gain of not using streams? That you don't > need to use fopen() yourself? (PHP will close files for you.) I'm just > not super convinced there's a need for adding this extra layer.
It's not about having to fopen() the stream – in my opinion at least that's a non-brainer. What is a minor issue though, is that either imagegifanimend() closes the stream automatically (what imagepng() and friends are doing), or not (in my opinion the proper solution, but inconsistent). The major issue for me would be the possibility that the user fiddles with the stream between imagegifanimbegin() and imagegifanimend() – in the worst case closing the stream, so imagegifanimend() would fail or had to catch that. Of course, that would be a programmer error, but nonetheless it would have to be documented, and I'm pretty sure that at least some devs would walk into this trap. Anyhow, thanks for the feedback. I'll see how to unify this in a single RFC (probably with secondary voting options).
-- Christoph M. Becker