tests/rm/rm-readdir-fail?: strace -o /dev/null -e inject=getdents64:error=EIO rm -rf has wrong error

[oech3]

$  mkdir -p a/b
$ strace -o /dev/null -e inject=getdents64:error=EIO rm -rf a
rm: cannot remove 'a': Directory not empty

$  mkdir -p a/b
$ strace -o /dev/null -e inject=getdents64:error=EIO gnu9.9/rm -rf a
rm: cannot remove 'a': Input/output error

[oech3]

@ChrisDryden do you think is this rm-readdir-fail? or something else?

[ChrisDryden]

There is an underlying issue here: #9604 this PR shows the fix to match the error message.

[oech3]

It seems you are patching test suite. Can we upstream strace instead?

[ChrisDryden]

That would be ideal, I was experimenting with this and the issue became that the upstream uses an artifact to prove that the intercepted method was actually called. Is there a way to do that with strace too?

[oech3]

I 1st considered #10272 (comment) , but it seems error message is different

[oech3]

Oh artifact... It is why coreutils/coreutils#153 (comment) was happened.

[ChrisDryden]

I was able to modify a similar test upstream when I found that depending on the version of GLIBC the test was incompatible when running with the original GNU coreutils, I think its likely that the same thing applies here and I can upstream the patch I wrote with sed

[oech3]

Great. (I don't want to sed too many soure code)