[P5P] Voting for cherry-picks for 5.20.2

On http://hmbrand.home.xs4all.nl/overview-5202.html I see

----------------------------------
(Note: #123339 is a prerequisite for #40565)
steveh 64c909d3055a7d5436012fab51eb1c046088140a 123339 fix segv for
psudofork duping of SAVEt_GP_ALIASED_SV
----------------------------------
Where did "(Note: #123339 is a prerequisite for #40565)" come from?

If you see
http://perl5.git.perl.org/perl.git/commit/64c909d3055a7d5436012fab51eb1c046088140a
you will see that the commit message refers to
http://perl5.git.perl.org/perl.git/commit/ff2a62e0c8 which is a commit
on Sept 19 2014 in blead. #123339 is a very similar bug I found in blead
while investigating #40565, but I dont think (if I am reading
http://perl5.git.perl.org/perl.git/commit/ff2a62e0c8 correct) it could
have existed in 5.20. #40565 has probably been around since day 1 of
psuedo fork, atleast since 2006 (I dont feel like testing my Strawberry
5.8 ATM). #123339 has been around since Sept 2014. The work up on
#123339 was done in #40565's ticket since I thought it was another
segving test case of the same bug due the SEGV callstack being nearly
identical until I figured out the problem and then the fix patch (save
stack type for #123339 is different from #40565's SS type)

http://perl5.git.perl.org/perl.git/commit/452466a503257b1012694b0d3a3cbf260e409d18
is needed though since its the perldelta entry for #40565 fix.

"Mathomise save_re_context"
http://perl5.git.perl.org/perl.git/commit/0ddd4a5b1910c8bfa9b7e55eb0db60a115fe368c
Absolutely no. Mathoming removes the function call from export table
(atleast on blead :p) on -DNO_MATHOMS builds so old XS modules compiled
against old headers and old libperl.a wont load into the process due to
missing symbol. Also according to P5P lore, changing the exported
symbols breaks binary compatibility since supposedly on some platforms
(16 bit Windows from 1985 cough cough) symbols are linked by numbers,
not strings. Resorting the def file would break linking then.

This is one of my notes in the original XML file. It was based on an
exchange between you and Father C in which you initially said 'yes and no'
to whether a proposed patch worked, and then FC asked whether it now worked
after your patch to 123339 had been committed and you said yes:

https://rt.perl.org/Public/Bug/Display.html?id=40565#txn-1321011

However, I hadn't realized that the 123339 problem was blead-specific,
given which I will of course drop it from the list. It's almost impossible
to tell exactly what's applicable and what isn't by just reading commit
logs and RT tickets. I tried to include everything that looked relevant
since it's easier to cross things out later than to pick up things that
I've missed, so there are bound to be some that turn out not to be relevant.
http://perl5.git.perl.org/perl.git/commit/64c909d3055a7d5436012fab51eb1c046088140a
you will see that the commit message refers to
http://perl5.git.perl.org/perl.git/commit/ff2a62e0c8 which is a commit on
Sept 19 2014 in blead. #123339 is a very similar bug I found in blead while
investigating #40565, but I dont think (if I am reading
http://perl5.git.perl.org/perl.git/commit/ff2a62e0c8 correct) it could have
existed in 5.20. #40565 has probably been around since day 1 of psuedo
fork, atleast since 2006 (I dont feel like testing my Strawberry 5.8 ATM).
#123339 has been around since Sept 2014. The work up on #123339 was done in
#40565's ticket since I thought it was another segving test case of the
same bug due the SEGV callstack being nearly identical until I figured out
the problem and then the fix patch (save stack type for #123339 is
different from #40565's SS type)
http://perl5.git.perl.org/perl.git/commit/452466a503257b1012694b0d3a3cbf260e409d18
is needed though since its the perldelta entry for #40565 fix.
http://perl5.git.perl.org/perl.git/commit/0ddd4a5b1910c8bfa9b7e55eb0db60a115fe368c
Absolutely no. Mathoming removes the function call from export table
(atleast on blead :p) on -DNO_MATHOMS builds so old XS modules compiled
against old headers and old libperl.a wont load into the process due to
missing symbol. Also according to P5P lore, changing the exported symbols
breaks binary compatibility since supposedly on some platforms (16 bit
Windows from 1985 cough cough) symbols are linked by numbers, not strings.
Resorting the def file would break linking then.
>

Thank you for the spot. I will drop this change from the list too. (XS
modules built against an old perl must indeed be able to load into a new
perl in the same maint series. I think *that* is precisely what people most
often mean when they speak of 'binary compatibility' in maint perl
releases.)

Please also consider updating Encode to 2.67 in 5.20.2, as it is
required to be able to get Tk running

(I haven't forgotten this request. I will certainly consider it, but I
just need to have a look at exactly what changes that would bring in.)

Thanks for this, Steve.

The attached patch tweaks the XSL so that the link for commits and
branches with three votes is shown in green. I'm finding this helps me
distinguish the changes that should still be considered from those
that are liable to be cherry-picked. (But I'm not applying it myself,
in case doing so makes your work as maint pumpking harder.)

The patch is hard to review, because the XSL is mostly on a single
line, so in summary:

- A <style> element is added that shows ".accepted .ref .a" in green

- The <td> containing the commit ID or branch name is given the CSS class ".ref"

- A <tr> for a commit with two or more commas in its "votes" attribute
is given the CSS class ".accepted"

I assume that the reason for putting the XSL on a single line is to
make it easier for committers to ignore it while they work on the
actual vote data. Would you like a patch that moves the XSL to the end
of the file and reindents it? I hope that would still make it easy to
vote on cherry-picking commits, while also simplifying the process of
changing the XSL if need be.

Oops, hit Reply instead of Reply All...


---------- Forwarded message ----------
From: Steve Hay <steve.m.hay@googlemail.com>
Date: 15 December 2014 at 18:30
Subject: Re: Voting for cherry-picks for 5.20.2
To: Aaron Crane <arc@cpan.org>

Thanks for voting and for your XSL improvement, which I've now pushed
to maint-5.20.

Yes, that was the reason for the ugly layout. I hadn't thought of
simply putting the XSL at the end! I've now done that, and made
another small improvement to display the section headers in the HTML
too.

Thanks — that's just the sort of thing I had in mind.

I've attached commits for three more proposed changes to the XSLT:

- Link to commitdiff pages (not just the commit message)
- Add a sub-heading for the "none of the above categories" commit
- Show commits with no votes at all in red

The last of those depends on a separate refactoring commit.

Thanks again for your work on 5.20.

Thanks, these look good to me. Now pushed to maint-5.20-votes.