[PATCH 2 of 2 v3] makefile: add wheel build target

[Gregory Szorc]

On Mon, Oct 12, 2015 at 11:42 AM, Gregory Szorc <gregory.szorc at gmail.com>
wrote:

> On Mon, Oct 12, 2015 at 7:39 AM, Augie Fackler <raf at durin42.com> wrote:
>
>> On Fri, Oct 09, 2015 at 12:31:16PM -0500, Nathan Goldbaum wrote:
>> > # HG changeset patch
>> > # User Nathan Goldbaum <ngoldbau at ucsc.edu>
>> > # Date 1444411846 18000
>> > #      Fri Oct 09 12:30:46 2015 -0500
>> > # Node ID c630785461f0465adebce581eb25a503f654333a
>> > # Parent  d27f31183f4ca4af2a0a96af32a631181b501b5a
>> > makefile: add wheel build target
>>
>> Queued. How well does the wheel work on Windows?
>>
>
> When I attempted this several months ago, I think it "just worked."
> However, I'm not sure hg.exe was produced. However, if you are installing
> Mercurial via wheels/pip on Windows, chances are you have a UNIX-like env
> (such as cygwin) and don't need the .exe wrapper.
>
> I can probably find time to investigate this on my Windows machine this
> week if there is any interest in producing wheels as part of the 3.6
> release.
>

`python.exe setup.py bdist_wheel` seems to produce a correct wheel,
compiled C extensions .pyd files and all! I did a `pip.exe install
mercurial-3.5.2-cp27-none-win_amd64.whl` into a virtualenv and it seemed to
"just work." `hg debuginstall` was happy. `hg serve` has access to
templates.

The one thing that isn't in the wheel is a hg.exe. But there is a hg.bat,
which is sufficient for `hg` to work in cmd.exe. As I mentioned in my
previous message, if you are installing a wheel, you probably have a
"proper" Python environment and don't need the standalone / Python-bundled
Mercurial that hg.exe/py2exe provides. So I don't think this is a problem.

Another issue might be the file naming. Even though I built @, the filename
and version metadata inside the wheel was "3.5.2" and didn't have the "+N"
component that we put in `hg --version`. Not sure if this is a bug or by
design.

I performed a `python.exe setup.py bdist_egg` and compared the resulting
zip archives. The content was exactly the same except a) the egg had an
EGG-INFO directory b) the wheel had mercurial-3.5.2.data and
mercurial-3.5.2.dist-info directory c) the egg had .pyc files. The .pyc
files made the egg ~2.6 MB instead of ~1.3 MB for the wheel. I /thought/
wheel files contained .pyc files. But I must be wrong or it isn't relevant
because `pip.exe install` appeared to create the .pyc files automagically.

I'm not a wheel expert and there might be some issues with the low-level
wheel metadata (content of mercurial-<version>.dist-info directory) but
AFAICT the produced .whl file looks like a valid, working Mercurial
distribution! I dare say "it just works."
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://selenic.com/pipermail/mercurial-devel/attachments/20151012/f46ba6c0/attachment.html>