Enable dumping output only if FILECHECK_DUMP_INPUT_ON_FAILURE is set to
a non-empty value. This is necessary to support disabling it via
POSIX-compliant env(1) that does not support '-u' argument,
and therefore fix regression caused by r366980.
Details
Diff Detail
Event Timeline
LGTM as is.
A few notes we should keep in mind anyway: FILECHECK_DUMP_INPUT_ON_FAILURE is deprecated, so I agree that this simple change is probably sufficient (we don't need to check for 0, for example). I wonder if anyone still uses it. If so, this patch might break some uses, but I think users could easily adjust. If not, we should consider removing it.
Thanks.
We are using it to force verbose output on buildbot ;-). But setting to an empty value should be very uncommon. The uses I've been able to grep in LLVM all set it to 1.
Good to know. By the way, you'd get roughly the same output with FILECHECK_OPTS=-dump-input=fail, which is not deprecated. Adding -v or -vv will provide much more info.
But setting to an empty value should be very uncommon.
Agreed.
The uses I've been able to grep in LLVM all set it to 1.
You mean the FileCheck test suite? Or are you grepping some user code?
Hm, indeed. However, when grepping help I've noticed FILECHECK_DUMP_INPUT_ON_FAILURE but not FILECHECK_OPTS ;-).
But setting to an empty value should be very uncommon.
Agreed.
The uses I've been able to grep in LLVM all set it to 1.
You mean the FileCheck test suite? Or are you grepping some user code?
The former.
It's in FileCheck.rst, but it should probably be in --help as well. I'll make a note to look into that. Thanks!