On 07/08/2021 22:58, Junio C Hamano wrote:
Noah Pendleton <noah.pendleton@xxxxxxxxx> writes:
That cuts both ways, though. Failing upon missing configuration
file is a way to catch misconfiguration that is hard to diagnose.
I wonder if we can easily learn where the configuration variable
came from in the codepath that diagnoses it as a misconfiguration.
If it came from a per-repo configuration and names a non-existent
file, it clearly is a misconfiguration that we want to flag as an
error. Even if it came from a per-user configuration, if it was
specified in a conditionally included file, it is likely to be a
misconfiguration. If it came from a per-user configuration that
applies without any condition, it can be a good convenience feature
to silently (or with a warning) ignore missing file.
I am very interested in this feature, but I'd like to add another point
to the discussion: in the case of ignoreRevsFile in particular, no one
creates a repository with such a file; it is always added later.
However, when bisecting (a typical usage scenario for git-blame), we may
end up returning back to a point _before_ the file had been added, and
then, git-blame fails. This often happens to me, and I am then forced to
`touch` the file to create it again, only to ensure git-blame keeps
working. And then, when I want to return to the HEAD commit, the file
must be erased again otherwise there is a conflict. So, for me, the
"ignore if absent" behavior seems to me like it should be the default.
