macOS Intel CI#135
Conversation
| if [ ! -f config.guess ]; then wget --continue https://raw.githubusercontent.com/gcc-mirror/gcc/master/config.guess; fi | ||
| if [ ! -f config.sub ]; then wget --continue https://raw.githubusercontent.com/gcc-mirror/gcc/master/config.sub; fi |
There was a problem hiding this comment.
we shouldn't use this mirror with old config files. I think we can do the same you did on ps3libraries repo, and add the get_config script, that tries to download but has a fallback.
Also we remove the duplicate download on gcc/binutils/ppu/spu and have a single download point.
There was a problem hiding this comment.
Agreed, thanks for the catch.
|
This PR can't proceed as long as ps3libraries is blocking. Changes to ps3libraries are currently affecting the reliability of the toolchain wholesale. Will reopen once it stabilises. |
|
The utime.h thing was already fixed in the newlib-1.20.0-PS3 repo, and the patch updated from that. Don't go making manual changes to |
That’s what I thought too, but I was hitting an issue with it during CI runs. Maybe I wasn’t rebasing, I’ll roll back and try it again today. |
|
It's still occurring on latest CI, https://github.com/ps3dev/ps3toolchain/actions/runs/27085703145/job/79939629234?pr=135 @zeldin do you know what the root cause of this is? |
|
@TheMrIron2 The root cause has been fixed already, but since your |
Replaced a bad path. On modern macOS, both `TARGET_OS_MAC` and `__APPLE__` can be visible. So the old "no fdopen()” path runs and the macro replaces fdopen() with NULL and corrupts it.
Same behaviour as ps3libraries - check Savannah, and if not, fall back on local copy
sys/utime.h declares utimbuf using time_t, but does not include a header that declares time_t. This breaks libraries that include <utime.h> directly, eg libnfs. Include <sys/types.h> earlier so the header is self-contained.
This reverts commit 9475164.
This reverts commit 93f2b6d.
|
BTW, this PR seems to contain two completely orthogonal things:
These should be split into two separate (and appropriately named) PRs before merge. |
|
That did the trick, Ubuntu CI is back to passing. Still getting the hang of this! A lot of moving parts here haha. Yes, one bled into the other as I worked on it, Savannah was hanging some CI attempts. I'll split them up once I verify macOS is clean. Thanks! |
|
Closing in favour of two separated PRs. |
Working macOS CI builds with GitHub actions.
Main changes:
macos-15-intelas stable target with correct stagingzlib)Messy commit history due to CI retesting so squash and merge if approved.