w3ctestlib branch a

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

w3ctestlib branch a

Geoffrey Sneddon-4
Hi,

https://bitbucket.org/gsnedders/w3ctestlib/branch/cleanup-a-1 has a
branch containing a whole load of fixes for w3ctestlib, which leads to
it discovering more tests in the repo than previously. It would be nice
to get this merged into w3ctestlib proper.

In short, significant changes are:

 a) Change how HTMLSource handles encoding, making it use the normal
HTML means of detecting encoding. This allows us to find all the UTF-16
encoded tests we were previously missing. (It does, however, oblige us
to use html5lib 0.99999999 or above, which has breaking changes in it.
Note, however, this causes a massive diff in the built output because it
makes all attributes stay in source order.)

 b) Remove a bunch of unused features (given the goal of getting rid of
the build system, we probably aren't going to ever start using them, and
as such they're just unnecessary complexity that make it harder to
understand the code!). Support for multiple groups in a testsuite is
gone, as is all the code for the old reftest manifest format.

 c) A whole load of changes to move us slightly more in line with PEP8,
to have some more normal Python code style.

 d) Turn it into a real Python package, and add some tests.

This also makes it possible to run the build system under PyPy (5.0.1 is
definitely required, maybe later?), which locally takes 3 minutes to
build the testsuite instead of 3:30.

/Geoffrey.

Reply | Threaded
Open this post in threaded view
|

Re: w3ctestlib branch a

fantasai
On 07/14/2016 11:31 PM, Geoffrey Sneddon wrote:
>
> Note, however, this causes a massive diff in the built output
> because it makes all attributes stay in source order.)

Yayyyyyyy!!!!!!! That was annoying.

>   b) Remove a bunch of unused features (given the goal of getting rid of
> the build system, we probably aren't going to ever start using them, and
> as such they're just unnecessary complexity that make it harder to
> understand the code!). Support for multiple groups in a testsuite is
> gone, as is all the code for the old reftest manifest format.

For the reftest manifest output? That was mainly built for Mozilla,
if they're not using it I guess that's okay.

~fantasai

Reply | Threaded
Open this post in threaded view
|

Re: w3ctestlib branch a

Geoffrey Sneddon-4
On Fri, Jul 15, 2016 at 6:32 PM, fantasai <[hidden email]> wrote:
> On 07/14/2016 11:31 PM, Geoffrey Sneddon wrote:
>>
>>
>> Note, however, this causes a massive diff in the built output
>> because it makes all attributes stay in source order.)
>
>
> Yayyyyyyy!!!!!!! That was annoying.

Hey, I managed to ship one release of html5lib which gave attributes
in *reverse document order* just to confuse everyone. :)

>
>>   b) Remove a bunch of unused features (given the goal of getting rid of
>> the build system, we probably aren't going to ever start using them, and
>> as such they're just unnecessary complexity that make it harder to
>> understand the code!). Support for multiple groups in a testsuite is
>> gone, as is all the code for the old reftest manifest format.
>
>
> For the reftest manifest output? That was mainly built for Mozilla,
> if they're not using it I guess that's okay.

No, for the input. We have a small amount of code to support adding
reftests given in a reftest manifest file, though it's never called
anywhere.

/Geoffrey

Reply | Threaded
Open this post in threaded view
|

Re: w3ctestlib branch a

Geoffrey Sneddon-4
In reply to this post by Geoffrey Sneddon-4
On 15/07/16 04:31, Geoffrey Sneddon wrote:

> Hi,
>
> https://bitbucket.org/gsnedders/w3ctestlib/branch/cleanup-a-1 has a
> branch containing a whole load of fixes for w3ctestlib, which leads to
> it discovering more tests in the repo than previously. It would be nice
> to get this merged into w3ctestlib proper.
>
> In short, significant changes are:
>
>  a) Change how HTMLSource handles encoding, making it use the normal
> HTML means of detecting encoding. This allows us to find all the UTF-16
> encoded tests we were previously missing. (It does, however, oblige us
> to use html5lib 0.99999999 or above, which has breaking changes in it.
> Note, however, this causes a massive diff in the built output because it
> makes all attributes stay in source order.)
>
>  b) Remove a bunch of unused features (given the goal of getting rid of
> the build system, we probably aren't going to ever start using them, and
> as such they're just unnecessary complexity that make it harder to
> understand the code!). Support for multiple groups in a testsuite is
> gone, as is all the code for the old reftest manifest format.
>
>  c) A whole load of changes to move us slightly more in line with PEP8,
> to have some more normal Python code style.
>
>  d) Turn it into a real Python package, and add some tests.
>
> This also makes it possible to run the build system under PyPy (5.0.1 is
> definitely required, maybe later?), which locally takes 3 minutes to
> build the testsuite instead of 3:30.

There's now also
<https://bitbucket.org/gsnedders/w3ctestlib/branch/supersuite> which
adds one commit on-top of that which does very little in and of itself,
but in combination with <https://github.com/w3c/csswg-test/pull/1100>
gets us building a testinfo.data file for the whole repo, per the
resolution at the last F2F.

I know that Mozilla especially would like to get this landed relatively
soon such that it's possible to run with two-way syncing for Servo,
though I realise you're at the TAG meeting this week (and presumably
travelling after!).

/Geoffrey