Hello Salvatore,

> After a short discussion with Paul, wouldn't that imply though that
> there is an soname bump needed? Do you know has upstream considered
> this and if/or why not? Is there enough assurance nobody (even outside
> Debian world) is using that symbol?

Those are all good questions, I should have done a better job at
explaining this, so let me try doing it now.

sergiodj@ did a lot of work investigating this as well, so most of the
things I'll be saying here are his findings (although if I say
anything wrong here, blame it on me).

The "curl_jmpenv" variable declaration was guarded by a "#ifdef
HAVE_SIGSETJMP", but the variable would only be used on "#ifdef
USE_ALARM_TIMEOUT".
For Debian, "HAVE_SIGSETJMP" is true but "USE_ALARM_TIMEOUT" is false,
this is because we use the threaded resolver.

This means that "curl_jmpenv" was dead code, and double checking for
mentions of "curl_jmpenv" on codesearch.d.n only returns packages
which bundles curl, nothing using it.

Considering the variable was exposed, but not used anywhere (in our
builds with threaded resolver), I don't think there was any possible
way dependencies could be making use of it in any meaningful way (this
is talking about things outside of our official repositories now).

It doesn't make sense for upstream to bump SONAME because the variable
can still be exposed depending on the build config, it's just that it
wasn't supposed to be exposed for threaded resolvers first of all.

The CVE that is being fixed by that change should not affect our
binaries since we build with the threaded resolver, but I ended up
making a decision to still apply the patch to safeguard even the
sources.

> These are just a couple of question trying to understand what
> potential question from release team members my come for your unblock
> request.

Thank you for reviewing this.

> p.s.: note it looks autopkgtest view for curl was still blocking it
> because cwltool has a flaky test (on armel).

Yeah, curl suffers quite a bit from these since tons of reverse-deps
use it to fetch resources over the network and that's always flaky
(not sure if it's the case with cwitool specifically, but I'm used to
this sort of thing by now).

Cheers,

-- 
Samuel Henrique <samueloph>

Reply via email to