Junio C Hamano [off-list ref] writes:
Coming back to the command line syntax for the new feature, if I had
to choose, I would say
git diff --no-index [-<options>] [--] <path> <path> <pathspec>
perhaps? As we never compare anything other than two things,...
Actually, I am not so sure about this anymore.
It is not entirely fair to say that
git diff --no-index [--options] A B C D
as comparing A and B while using C and D as pathspec and declare a
person who expects otherwise a moron. To a person whose brain is
not rotten by years of use of Git, "we never compare anything other
than two things with --no-index" is not a given, I am afraid.
It is equally plausible that the same command line may be asking to
compare A with B and C with D, i.e. producing
diff --no-git a/A b/B
...
diff --no-git a/C b/D
...
It also equally plausible (taking a cue from "mv A B C Dir/") that
it is comparing A, B and C with D/A, D/B and D/C, respectively,
producing
diff --no-git a/A b/D/A
...
diff --no-git a/B b/D/B
...
diff --no-git a/C b/D/C
...
The only unambiguous syntax I can think of that avoids such
alternative interpretations is something ugly like
git diff --no-index [-<options>] --src=<path> --dst=<path> [--] <pathspec>
where "src" and "dst" are in line with the existing src/dst-prefix
options.
Perhaps we could declare that this is the "canonical" way to spell
pathspec-filtered no-index comparison of two directories, but we
allow the syntax suggested in the message I am responding to as a
short-hand, but I am not sure if that would fly well.
This --src/--dst thing could lead to even uglier tangent. We could
use these options to specify what is compared with what else
--{src,dst}-path=<directory in the working tree>
--{src,dst}-index
--{src,dst}-worktree
--{src,dst}-tree=<tree object>
which allows us to express funky combinations like
git diff --src-index --dst-worktree [--] <pathspec>
git diff --src-tree=HEAD --dst-index [--] <pathspec>
git diff --src-tree=maint --dst-tree=master [--] <pathspec>
The above three would be ugly ways to spell the traditional "short
hands", e.g.
git diff [--] <pathspec>
git diff --cached [--] <pathspec>
git diff maint master [--] <pathspec>
respectively. And an obvious fallout of the above is this command
line:
git diff --src-path=A --dst-path=B [--] <pathspec>
Because this does not involve any index, tree or worktree, it cannot
be a Git operation, so we know that is --no-index mode; the command
line does not need to say --no-index anywhere if we go that route.
On Sun, Sep 21, 2014 at 1:19 PM, Junio C Hamano [off-list ref] wrote:
Junio C Hamano [off-list ref] writes:
quoted
Coming back to the command line syntax for the new feature, if I had
to choose, I would say
git diff --no-index [-<options>] [--] <path> <path> <pathspec>
perhaps? As we never compare anything other than two things,...
Actually, I am not so sure about this anymore.
It is not entirely fair to say that
git diff --no-index [--options] A B C D
as comparing A and B while using C and D as pathspec and declare a
person who expects otherwise a moron. To a person whose brain is
not rotten by years of use of Git, "we never compare anything other
than two things with --no-index" is not a given, I am afraid.
It is equally plausible that the same command line may be asking to
compare A with B and C with D, i.e. producing
diff --no-git a/A b/B
...
diff --no-git a/C b/D
...
Heh.. I wrote something like that [1]. An extension of this is feed
all filename pairs via stdin.
[1] http://thread.gmane.org/gmane.comp.version-control.git/188435
It also equally plausible (taking a cue from "mv A B C Dir/") that
it is comparing A, B and C with D/A, D/B and D/C, respectively,
producing
diff --no-git a/A b/D/A
...
diff --no-git a/B b/D/B
...
diff --no-git a/C b/D/C
...
This mode is weird... In real world, "A" may match to D/some-path/A
not just D/A. Which brings it back to the previous mode.
The only unambiguous syntax I can think of that avoids such
alternative interpretations is something ugly like
git diff --no-index [-<options>] --src=<path> --dst=<path> [--] <pathspec>
where "src" and "dst" are in line with the existing src/dst-prefix
options.
Or let the user define a "separator", e.g.
git diff --no-index --pathspec-separator=<sep> [--] <paths...> [<sep>
<pathspec>]
Another option is pathspec magic (I'm not entirely sure how it looks
like at the implementaiton level yet), --src=<path> could become
:(source)path, --dst=path -> :(dst)path and so on..
Perhaps we could declare that this is the "canonical" way to spell
pathspec-filtered no-index comparison of two directories, but we
allow the syntax suggested in the message I am responding to as a
short-hand, but I am not sure if that would fly well.
This --src/--dst thing could lead to even uglier tangent. We could
use these options to specify what is compared with what else
--{src,dst}-path=<directory in the working tree>
--{src,dst}-index
--{src,dst}-worktree
--{src,dst}-tree=<tree object>
which allows us to express funky combinations like
git diff --src-index --dst-worktree [--] <pathspec>
git diff --src-tree=HEAD --dst-index [--] <pathspec>
git diff --src-tree=maint --dst-tree=master [--] <pathspec>
The above three would be ugly ways to spell the traditional "short
hands", e.g.
git diff [--] <pathspec>
git diff --cached [--] <pathspec>
git diff maint master [--] <pathspec>
respectively. And an obvious fallout of the above is this command
line:
git diff --src-path=A --dst-path=B [--] <pathspec>
Because this does not involve any index, tree or worktree, it cannot
be a Git operation, so we know that is --no-index mode; the command
line does not need to say --no-index anywhere if we go that route.
Sounds a lot like Jeff's "git-put", except with -- syntax instead of a
special one, and you only look at the diff, not move any content. Also
people who would go with "git diff --no-index A B C D E F..." usually
script it, not type it manually. And stdin would fit better than
command line arguments.
--
Duy