Repository navigation
Error during build on unguarded use of new symbols on macOS #128135
Description
Feature or enhancement
Proposal:
CPython builds run against the latest macOS version and unguarded use of new symbols does not throw an error. However, building CPython on older versions of macOS is supported. Consequently, regressions are not caught, e.g.:
- HACL Blake2 implementation does not check for
aligned_allocavailability on macOS #128134 - Availability of
ptsname_ris not checked at runtime on macOS #123797 - Availability of utimensat, futimens not checked correctly on macOS #75782
Enabling -Werror=unguarded-availability-new would be great to catch these before they become a problem downstream.
Has this already been discussed elsewhere?
This is a minor feature, which does not need previous discussion elsewhere
Links to previous discussion of this feature:
This was discussed briefly in Discord
Activity
[-]Error during build on unguarded use of new symbols[/-][+]Error during build on unguarded use of new symbols on macOS[/+]zanieb commentedon Dec 20, 2024
Here's an initial patch 36c7da4
I'm not sure if we should be using
unguarded-availabilityorunguarded-availability-newand define an explicit lower bound version? We use the latter downstream, but we also set a minimum version.I'm having a hard time reproducing failures without cross-compiling via
python-build-standalone. Probably user error.zanieb commentedon Dec 20, 2024
Another complexity is astral-sh/python-build-standalone#210 / astral-sh/python-build-standalone#422 — if this is in the
CFLAGSinsysconfigit is used to build extension modules in some packages which causes failures.zanieb commentedon Dec 20, 2024
Perhaps this is dumb to ask, but.. this flag is clang only? I don't think there's an obvious equivalent in gcc?
picnixz commentedon Dec 20, 2024
I'm not aware of an equivalent one for gcc. Btw, I think we already had an issue for this one: #100384 but I think it's better to keep the latest in this case unless you want to move the discussion out there.
zanieb commentedon Dec 20, 2024
Ah thank you @picnixz — I don't have a preference for which issue we use but agree it makes sense to close one of them.
picnixz commentedon Dec 20, 2024
Let's move the discussion on the original one. I'll edit the issue to include your findings.