-Csave-temps=on doesn't actually save anything

[infinity0]

#42864 doesn't work. I need it to debug a linker failure, to re-run the linker command. But the tempdir isn't there even when I pass -Csave-temps=on.

@slo1 @Mark-Simulacrum

[infinity0]

Work around:

$ cat binhack/x86_64-w64-mingw32-gcc
#!/bin/sh
/usr/bin/x86_64-w64-mingw32-gcc "$@"
x=$?
mkdir -p /tmp/savedrustc/
cp -a /tmp/rustc* /tmp/savedrustc/
exit $x

$ PATH=$PWD/binhack:$PATH sh test.sh
[..]

$ $ ls -gG /tmp/savedrustc/rustc34QKlf/
total 3388
-rw-r--r-- 1  233304 Aug  8 03:31 liballoc-62d18a8976d7c133.rlib
-rw-r--r-- 1  135158 Aug  8 03:31 libbacktrace-4ceb221c1451fd72.rlib
-rw-r--r-- 1  584318 Aug  8 03:31 libbacktrace_sys-5ac9ad0bd2943e4c.rlib
-rw-r--r-- 1    2150 Aug  8 03:31 libcfg_if-b787a80b1cca0552.rlib
-rw-r--r-- 1  484624 Aug  8 03:31 libcompiler_builtins-fa2917f0b54acfc4.rlib
-rw-r--r-- 1 1547472 Aug  8 03:31 libcore-bae6c7eaa8cff910.rlib
-rw-r--r-- 1   20156 Aug  8 03:31 libhashbrown-951231d947810b4e.rlib
-rw-r--r-- 1    9532 Aug  8 03:31 liblibc-3072b25f7851b3d1.rlib
-rw-r--r-- 1   38166 Aug  8 03:31 libpanic_unwind-9c93af3a21bf779f.rlib
-rw-r--r-- 1  182730 Aug  8 03:31 librustc_demangle-176970e988fbc465.rlib
-rw-r--r-- 1    2242 Aug  8 03:31 librustc_std_workspace_alloc-4397ecfaeb9d832e.rlib
-rw-r--r-- 1    2244 Aug  8 03:31 librustc_std_workspace_core-1b8ab73855770e4f.rlib
-rw-r--r-- 1    6702 Aug  8 03:31 libunwind-204b2abdb5f7ae9a.rlib
-rw-r--r-- 1  199457 Aug  8 03:31 list

[Mark-Simulacrum]

Hm I am seeing the directory stick around at least, though on a trivial hello world file it is empty. I'm not sure yet if that's expected...

[infinity0]

I also tried a hello-world example but when passing -Clink-arg=xxx to force a linker failure, didn't notice any "/tmp" paths in the command output, so assumed it didn't need temp files. You can probably trigger it by running RUSTFLAGS=-Clink-arg=xxx ./x.py build or something.

[infinity0]

Ah, the problem possibly is that the directory is only saved when rustc doesn't fail.

[Mark-Simulacrum]

Hopefully this'll be fixed by #75315.