git silently ignores aliases of existing commands

> On Fri, 17 Jul 2009 17:52:49 -0700
> Michael G Schwern <schwern@pobox.com> wrote:
> 
> [...]
>> It would be nice if git used the alias *before* the installed command.  This
>> lets me fix/change default behaviors without having to come up with a new
>> command.  (Another handy example:  blame = blame -w)  It doesn't do anything
>> useful right now anyway.
> 
> This has been discussed a few times on the list already.   Here is one such
> discussion:
> 
> http://thread.gmane.org/gmane.comp.version-control.git/112487/focus=112493
> 
> You'll see that it was decided that Git would not allow commands to be overridden
> so that you could always be sure what a given command would do when you sit
> down at any installation.  This is especially important for scripting but can
> also be a problem for everyday usage.   You'll just have to choose a new command
> name for the alternate default you want.

> Sean Estabrooks wrote:
>> On Fri, 17 Jul 2009 17:52:49 -0700
>> Michael G Schwern <schwern@pobox.com> wrote:
>>
>> [...]
>>> It would be nice if git used the alias *before* the installed command.  This
>>> lets me fix/change default behaviors without having to come up with a new
>>> command.  (Another handy example:  blame = blame -w)  It doesn't do anything
>>> useful right now anyway.
>>
>> This has been discussed a few times on the list already.   Here is one such
>> discussion:
>>
>> http://thread.gmane.org/gmane.comp.version-control.git/112487/focus=112493
>>
>> You'll see that it was decided that Git would not allow commands to be overridden
>> so that you could always be sure what a given command would do when you sit
>> down at any installation.  This is especially important for scripting but can
>> also be a problem for everyday usage.   You'll just have to choose a new command
>> name for the alternate default you want.
>
> I'm in the "more than enough rope" camp myself, so count that as a -1 fwiw.
>
> More importantly, what about the warning telling the user that what they did
> is not allowed and didn't work?

> On Sat, Jul 18, 2009 at 11:30:25AM +0200, demerphq wrote:
>
>> Yeah it seems reasonable that if its going to be ignored it should not
>> be silently ignored.
>
> I agree that there should be a warning. However, it's hard to do with
> the code structured as it is now; we don't know that a command exists
> as an external command until we try to exec it. And if it succeeds, we
> don't get to execute any more code.
>
> It's certainly possible, but sadly it is more surgery than just
>
>  if (alias_lookup(cmd))
>    warn("you also have an alias defined");
>
>> Especially given that the silentness effectively means there cant be
>> any new git tools added without possible breakage of installed setups.
>
> The silentness makes it harder to diagnose problems, but even with a
> warning, we can break things by creating new commands. If you have an
> alias "foo" and we ship "git-foo" in a newer version of git, your alias
> will just stop working.

> > The silentness makes it harder to diagnose problems, but even with a
> > warning, we can break things by creating new commands. If you have an
> > alias "foo" and we ship "git-foo" in a newer version of git, your alias
> > will just stop working.
> 
> That was my point. At least if there were warnings about this the risk
> would be mitigated.

> On Sat, Jul 18, 2009 at 12:55:26PM +0200, demerphq wrote:
>
>> > The silentness makes it harder to diagnose problems, but even with a
>> > warning, we can break things by creating new commands. If you have an
>> > alias "foo" and we ship "git-foo" in a newer version of git, your alias
>> > will just stop working.
>>
>> That was my point. At least if there were warnings about this the risk
>> would be mitigated.
>
> I don't see how it's mitigated. You don't get any warning until _after_
> things are broken. So yes, it may help you diagnose the breakage, but
> presumably the fact that the command is doing something completely
> different would also alert you to the breakage.
>
> The real problem comes from scripted use, where you don't necessarily
> have a user reading warnings on stderr, or notice that some totally
> bogus code is being run (especially if said code happens not to produce
> a non-zero exit code).
>
> But perhaps that's what you meant, and I'm just nitpicking your
> language.

> 2009/7/18 Jeff King <peff@peff.net>:
>> On Sat, Jul 18, 2009 at 12:55:26PM +0200, demerphq wrote:
>>
>>>> The silentness makes it harder to diagnose problems, but even with a
>>>> warning, we can break things by creating new commands. If you have an
>>>> alias "foo" and we ship "git-foo" in a newer version of git, your alias
>>>> will just stop working.
>>> That was my point. At least if there were warnings about this the risk
>>> would be mitigated.
>> I don't see how it's mitigated. You don't get any warning until _after_
>> things are broken. So yes, it may help you diagnose the breakage, but
>> presumably the fact that the command is doing something completely
>> different would also alert you to the breakage.
>>
>> The real problem comes from scripted use, where you don't necessarily
>> have a user reading warnings on stderr, or notice that some totally
>> bogus code is being run (especially if said code happens not to produce
>> a non-zero exit code).
>>
>> But perhaps that's what you meant, and I'm just nitpicking your
>> language.
> 
> I think we are more or less in agreement, except maybe that i think
> the situation would be marginally better if git detected this.
> 
> :-)
> 
> Seems an awkward position actually. Maybe a switch like
> --ignore-command-aliases which would be used by all internal commands
> when they expect to find another internal command would resolve it.
> Then aliases of internal commands to control default switches could
> actually be allowed to work, and there would not be the future
> compatibility trap that there seems to be now.