On Sat, Sep 08 2018, Stas Bekman wrote:
> On 2018-09-08 12:54 PM, Ævar Arnfjörð Bjarmason wrote:
>>
>> On Sat, Sep 08 2018, Martin Ågren wrote:
>>
>>> Hi Stas
>>>
>>> On Sat, 8 Sep 2018 at 21:00, Stas Bekman <stas@xxxxxxxxxx> wrote:
>>>> [include]
>>>> path = '../.gitconfig'
>>>>
>>>> Notice the single quotes around the filename. When this is the case git
>>>> silently (!) ignores the custom configuration, which is clearly a bug.
>>>
>>> Thanks for reporting and describing out your expectations and what you
>>> observed.
>>>
>>> Actually, there is a test explicitly testing that 'missing include files
>>> are ignored'. I couldn't find a motivation for this in 9b25a0b52e
>>> (config: add include directive, 2012-02-06).
>>>
>>>> The original problem cropped up due to using:
>>>>
>>>> git config --local include.path '../.gitconfig'
>>>>
>>>> which on linux stripped the single quotes, but on some windows git bash
>>>> emulation it kept them.
>>>
>>> Huh, I wouldn't have expected them to be kept. You learn something
>>> new every day...
>>>
>>>> What am I suggesting is that git:
>>>>
>>>> (1) should complain if it encounters an invalid configuration and not
>>>> silently ignore it. It took quite some effort and time to figure the
>>>> culprit.
>>>
>>> Sounds reasonable to me, but I might be missing something. I'm cc-ing
>>> the original author. Maybe he can recall why he made sure it silently
>>> ignores missing files.
>>>
>>>> (2) probably allow the quoted location of the file, but it's much less
>>>> important, as it's easy to rectify once git gives user #1
>>>
>>> I don't think this will work. Allowing quoting for just this one item,
>>> or for all? Any and all quoting or just at the first and last character?
>>> What about those config items where quotes might legitimately occur,
>>> i.e., we'd need some escaping? Actually, something like '.gitconfig'
>>> *with* *those* *quotes* is a valid filename on my machine.
>>
>> The reason missing includes are ignored is that the way this is expected
>> to be used is e.g.:
>>
>> [include]
>> path ~/.gitconfig.work
>>
>> Where .gitconfig.work is some configuration you're going to drop into
>> place on your $dayjob servers, but not on your personal machine, even
>> though you sync the same ~/.gitconfig everywhere.
>
> Thank you for clarifying why this is done silently, Ævar. It makes sense
> then.
>
>> If we were to make nonexisting files an error, we'd need something like
>> an extension of the includeIf syntax added in 3efd0bedc6 ("config: add
>> conditional include", 2017-03-01) 3efd0bedc6 ("config: add conditional
>> include", 2017-03-01). I.e.:
>>
>> [includeIfcond "test -e ~/.gitconfig.work"]
>> path = ~/.gitconfig.work
>>
>> Or something like that, this is getting increasingly harder to shove
>> into the *.ini config syntax.
>
> This suggestion won't solve the real problem. The real problem is that
> git can't find '.gitconfig' even though it's there, due to single quotes
> around the filepath. So the suggested check will still ignore the
> configuration even if it's there.
...because that's not how the *.ini syntax works. That means to look up
a file called '.gitconfig', as opposed to .gitconfig, ie. one that
actually starts with a single quote. On POSIX systems filenames can
include all bytes except \0, so we need some way to include those.
I've just created a 'foo' file (i.e. one that has a 5-chararcer name,
including single quotes), and including it via git's config works, as
opposed to the filename foo (i.e. the three-character version).
I can see how this is confusing, but we can't have some way to have this
"ignore missing" feature and warn about stuff like 'foo' v.s. "foo"
v.s. foo without carrying some list of quoting constructs deemed to be
confusing, and forbidding includes from files that look like that.
> This also leads me to think what if the include path has spaces in it?
>
> path = ~/somewhere on my system/.gitconfig.work
>
> most people would assume quotes are needed around the filepath.
