diff mbox series

Git alias syntax help

Message ID CAHMHMxWpLAnj3w8DGLMFbfy-A-pBjDxNdMeiM-fyuu-gnZyg+Q@mail.gmail.com (mailing list archive)
State New, archived
Headers show
Series Git alias syntax help | expand

Commit Message

Σταύρος Ντέντος Jan. 14, 2020, 3:21 p.m. UTC
Hello there,

I am having an issue with git-aliases - specifically, the intricacies
involved in their syntax.

In general, the syntax is confusing to me, especially when it is
_wise_ to use quotes inside a `!sh` alias.
e.g. which one would be the correct one
new = "!f() { : git log ; git log \"${1}@{1}..${1}@{0}\" \"$@\" ; } ; f"
new = !f() { : git log ; git log "${1}@{1}..${1}@{0}" "$@" ; } ; f

(from converting this:
https://git.wiki.kernel.org/index.php/Aliases#What.27s_new.3F)
        new = !sh -c 'git log $1@{1}..$1@{0} "$@"'

The alias confusing me is more specifically this:
https://git.wiki.kernel.org/index.php/Aliases#simple_diff_ignoring_line_number_changes

diffsort = !sh -c 'git diff "$@" | grep "^[+-]" | sort --key=1.2 | uniq -u -s1'

The output of:
$  colordiff -su <(git diffsort HEAD^..HEAD) <(git diffsort HEAD^^..HEAD^)
Files /dev/fd/63 and /dev/fd/62 are identical
is a little unexpected, since I know for a fact that one of the
referced commits is not a code block moved.

(and indeed, if I do it myself:
$ colordiff -su <(git diff HEAD^..HEAD | grep "^[+-]" | sort --key=1.2
| uniq -u -s1) <(git diff HEAD^^..HEAD^ | grep "^[+-]" | sort
--key=1.2 | uniq -u -s1)
I would appreciate if someone would clear out the air for me.
I think I have done my homework enough, and it is not trivially
obvious what is the correct thing to do.

I'll gladly take pointers though

Ντέντος Σταύρος

Comments

Jeff King Jan. 14, 2020, 10:28 p.m. UTC | #1
On Tue, Jan 14, 2020 at 05:21:20PM +0200, Σταύρος Ντέντος wrote:

> I am having an issue with git-aliases - specifically, the intricacies
> involved in their syntax.
> 
> In general, the syntax is confusing to me, especially when it is
> _wise_ to use quotes inside a `!sh` alias.
> e.g. which one would be the correct one
> new = "!f() { : git log ; git log \"${1}@{1}..${1}@{0}\" \"$@\" ; } ; f"
> new = !f() { : git log ; git log "${1}@{1}..${1}@{0}" "$@" ; } ; f

Only the first one is correct. In addition to the quotes in the second
one being eaten by the config parser, the unquoted semicolon starts a
comment.

> The alias confusing me is more specifically this:
> https://git.wiki.kernel.org/index.php/Aliases#simple_diff_ignoring_line_number_changes
> 
> diffsort = !sh -c 'git diff "$@" | grep "^[+-]" | sort --key=1.2 | uniq -u -s1'
> 
> The output of:
> $  colordiff -su <(git diffsort HEAD^..HEAD) <(git diffsort HEAD^^..HEAD^)
> Files /dev/fd/63 and /dev/fd/62 are identical
> is a little unexpected, since I know for a fact that one of the
> referced commits is not a code block moved.

The issue here isn't with Git's alias mechanism, but a quirk of how "sh
-c" works.  You can run with GIT_TRACE to see what we're passing to the
shell (though note that your double-quotes don't make it through):

  $ GIT_TRACE=1 git diffsort HEAD^..HEAD
  17:22:47.644542 [pid=3959333] git.c:708           trace: exec: git-diffsort HEAD^..HEAD
  17:22:47.644648 [pid=3959333] run-command.c:663   trace: run_command: git-diffsort HEAD^..HEAD
  17:22:47.645038 [pid=3959333] run-command.c:663   trace: run_command: 'sh -c '\''git diff $@ | grep ^[+-] | sort --key=1.2 | uniq -u -s1'\''' HEAD^..HEAD
  17:22:47.650319 [pid=3959336] git.c:439           trace: built-in: git diff

The problem is that "sh -c" takes the first non-option argument as $0,
not $1. For example:

  $ sh -c 'echo 0=$0, @=$@' foo bar baz
  0=foo, @=bar baz

You can add any extra string there to become $0, like:

  diffsort = "!sh -c 'git diff \"$@\" | grep \"^[+-]\" | sort --key=1.2 | uniq -u -s1' --"

which will do what you want. You can use whatever string you like, since
you know that your "-c" snippet does not ever look at $0.

-Peff
Σταύρος Ντέντος Jan. 14, 2020, 11:13 p.m. UTC | #2
On Wed, 15 Jan 2020 at 00:28, Jeff King <peff@peff.net> wrote:
>
> On Tue, Jan 14, 2020 at 05:21:20PM +0200, Σταύρος Ντέντος wrote:
>
> > I am having an issue with git-aliases - specifically, the intricacies
> > involved in their syntax.
> >
> > In general, the syntax is confusing to me, especially when it is
> > _wise_ to use quotes inside a `!sh` alias.
> > e.g. which one would be the correct one
> > new = "!f() { : git log ; git log \"${1}@{1}..${1}@{0}\" \"$@\" ; } ; f"
> > new = !f() { : git log ; git log "${1}@{1}..${1}@{0}" "$@" ; } ; f
>
> Only the first one is correct. In addition to the quotes in the second
> one being eaten by the config parser, the unquoted semicolon starts a
> comment.

Could somehow the latter "become" the correct option?
Especially in the case of `!sh`:
1) You need to quote everything after `=` sign ("forced" double quotes), then
2) `sh -c` needs another set (singles are most safe here, I think), and
3) If, for some reason, you need to quote further ("$@" would be a
common suspect usually)

Apart from the [1] feeling unneeded (the equivalent of Python's
`alias_cmd = cfg_line.split()[1]` could be enough), this brings a
quoting mess on [3]

> > The alias confusing me is more specifically this:
> > https://git.wiki.kernel.org/index.php/Aliases#simple_diff_ignoring_line_number_changes
> >
> > diffsort = !sh -c 'git diff "$@" | grep "^[+-]" | sort --key=1.2 | uniq -u -s1'
> >
> > The output of:
> > $  colordiff -su <(git diffsort HEAD^..HEAD) <(git diffsort HEAD^^..HEAD^)
> > Files /dev/fd/63 and /dev/fd/62 are identical
> > is a little unexpected, since I know for a fact that one of the
> > referced commits is not a code block moved.
>
> The issue here isn't with Git's alias mechanism, but a quirk of how "sh
> -c" works.  You can run with GIT_TRACE to see what we're passing to the
> shell (though note that your double-quotes don't make it through):
>
>   $ GIT_TRACE=1 git diffsort HEAD^..HEAD
>   17:22:47.644542 [pid=3959333] git.c:708           trace: exec: git-diffsort HEAD^..HEAD
>   17:22:47.644648 [pid=3959333] run-command.c:663   trace: run_command: git-diffsort HEAD^..HEAD
>   17:22:47.645038 [pid=3959333] run-command.c:663   trace: run_command: 'sh -c '\''git diff $@ | grep ^[+-] | sort --key=1.2 | uniq -u -s1'\''' HEAD^..HEAD
>   17:22:47.650319 [pid=3959336] git.c:439           trace: built-in: git diff
>
> The problem is that "sh -c" takes the first non-option argument as $0,
> not $1. For example:
>
>   $ sh -c 'echo 0=$0, @=$@' foo bar baz
>   0=foo, @=bar baz
>
> You can add any extra string there to become $0, like:
>
>   diffsort = "!sh -c 'git diff \"$@\" | grep \"^[+-]\" | sort --key=1.2 | uniq -u -s1' --"
>
> which will do what you want. You can use whatever string you like, since
> you know that your "-c" snippet does not ever look at $0.
>
> -Peff

Thank you very much for a complete explaination of all of this .

Can some of this be documented somewhere?
Are they somewhere and I missed them?

If nothing more, a link to this e-mail chain either on the wiki (if
https://git.wiki.kernel.org/index.php/Aliases is an official page) or
on git-alias help (here
https://git-scm.com/book/en/v2/Git-Basics-Git-Aliases or in some
"advanced" section, which I cannot find)

If https://git.wiki.kernel.org/index.php/Aliases is an official page,
then: was this written for an earlier version?
Could it also be updated?

--
Ntentos Stavros
Jeff King Jan. 15, 2020, 6:13 p.m. UTC | #3
On Wed, Jan 15, 2020 at 01:13:17AM +0200, Σταύρος Ντέντος wrote:

> > > new = "!f() { : git log ; git log \"${1}@{1}..${1}@{0}\" \"$@\" ; } ; f"
> > > new = !f() { : git log ; git log "${1}@{1}..${1}@{0}" "$@" ; } ; f
> >
> > Only the first one is correct. In addition to the quotes in the second
> > one being eaten by the config parser, the unquoted semicolon starts a
> > comment.
> 
> Could somehow the latter "become" the correct option?

Unfortunately not without breaking compatibility of existing config
files.

> Especially in the case of `!sh`:
> 1) You need to quote everything after `=` sign ("forced" double quotes), then
> 2) `sh -c` needs another set (singles are most safe here, I think), and
> 3) If, for some reason, you need to quote further ("$@" would be a
> common suspect usually)

Yes, the quoting can get pretty hairy. If your command is complicated, I
suggest writing it as a separate script and dropping it into your $PATH
as git-new. Then "git new" will run it automatically (and it's not even
any less efficient; it still ends up with a single shell invocation).

> Thank you very much for a complete explaination of all of this .
> 
> Can some of this be documented somewhere?
> Are they somewhere and I missed them?

I think the config syntax around quoting is described in "git help
config" (see the section Syntax). The shell parts seem out of scope for
Git's manpages themselves, though it sounds like maybe some examples you
found could stand to be fixed (and/or to call out the subtlety).

> If nothing more, a link to this e-mail chain either on the wiki (if
> https://git.wiki.kernel.org/index.php/Aliases is an official page) or
> on git-alias help (here
> https://git-scm.com/book/en/v2/Git-Basics-Git-Aliases or in some
> "advanced" section, which I cannot find)
> 
> If https://git.wiki.kernel.org/index.php/Aliases is an official page,
> then: was this written for an earlier version?
> Could it also be updated?

I think both of the issues you mentioned have always been true. That
wiki is open to editing by the world, so it's possible that somebody
just added bad examples (and fixes would be welcome).

I don't know the book content very well. Looking at the page you
linked, I don't think it says anything _wrong_, but it definitely
doesn't discuss more advanced alias usage. I suspect that would be a
welcome addition; they take contributions at
https://github.com/progit/progit2.

-Peff
Philip Oakley March 28, 2020, 5:05 p.m. UTC | #4
Hi,

On 15/01/2020 18:13, Jeff King wrote:
> On Wed, Jan 15, 2020 at 01:13:17AM +0200, Σταύρος Ντέντος wrote:
>
>>>> new = "!f() { : git log ; git log \"${1}@{1}..${1}@{0}\" \"$@\" ; } ; f"
>>>> new = !f() { : git log ; git log "${1}@{1}..${1}@{0}" "$@" ; } ; f
>>> Only the first one is correct. In addition to the quotes in the second
>>> one being eaten by the config parser, the unquoted semicolon starts a
>>> comment.
>> Could somehow the latter "become" the correct option?
> Unfortunately not without breaking compatibility of existing config
> files.
>
>> Especially in the case of `!sh`:
>> 1) You need to quote everything after `=` sign ("forced" double quotes), then
>> 2) `sh -c` needs another set (singles are most safe here, I think), and
>> 3) If, for some reason, you need to quote further ("$@" would be a
>> common suspect usually)
> Yes, the quoting can get pretty hairy. If your command is complicated, I
> suggest writing it as a separate script and dropping it into your $PATH
> as git-new. Then "git new" will run it automatically (and it's not even
> any less efficient; it still ends up with a single shell invocation).
>
>> Thank you very much for a complete explaination of all of this .
>>
>> Can some of this be documented somewhere?
>> Are they somewhere and I missed them?
> I think the config syntax around quoting is described in "git help
> config" (see the section Syntax). The shell parts seem out of scope for
> Git's manpages themselves, though it sounds like maybe some examples you
> found could stand to be fixed (and/or to call out the subtlety).
>
>> If nothing more, a link to this e-mail chain either on the wiki (if
>> https://git.wiki.kernel.org/index.php/Aliases is an official page) or
>> on git-alias help (here
>> https://git-scm.com/book/en/v2/Git-Basics-Git-Aliases or in some
>> "advanced" section, which I cannot find)
>>
>> If https://git.wiki.kernel.org/index.php/Aliases is an official page,
>> then: was this written for an earlier version?
>> Could it also be updated?
> I think both of the issues you mentioned have always been true. That
> wiki is open to editing by the world,

I took the liberty of adding a small subsection on quoting based on this
discussion, and adding a link to it. Feedback on my ignorance welcomed ;-)
>  so it's possible that somebody
> just added bad examples (and fixes would be welcome).
>
> I don't know the book content very well. Looking at the page you
> linked, I don't think it says anything _wrong_, but it definitely
> doesn't discuss more advanced alias usage. I suspect that would be a
> welcome addition; they take contributions at
> https://github.com/progit/progit2.
>
> -Peff
Philip
diff mbox series

Patch

--- /dev/fd/63 2020-01-14 17:17:45.103771745 +0200
+++ /dev/fd/62 2020-01-14 17:17:45.103771745 +0200
@@ -1,2 +1,13 @@ 
[....]

it works.

The issue I have found is: $@ is not expanded

Tested with:

diffsort = !sh -c 'echo "+git diff $@" | grep "^[+-]" | sort --key=1.2
| uniq -u -s1'
$ git diffsort HEAD^..HEAD
+git diff