[PATCH] doc: filter-branch does not require re-export of vars

[PATCH] doc: filter-branch does not require re-export of vars

[Andreas Heiduk]

The function `set_ident` in `filter-branch` exported the variables
GIT_(AUTHOR|COMMITTER)_(NAME|EMAIL|DATE) at least since 6f6826c52b in 2007.
Therefore the filter scripts don't need to re-eport them again.

Signed-off-by: Andreas Heiduk <asheiduk@gmail.com>
---
 Documentation/git-filter-branch.txt | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/Documentation/git-filter-branch.txt b/Documentation/git-filter-branch.txt
index 6e4bb0220..7b695dbb7 100644
--- a/Documentation/git-filter-branch.txt
+++ b/Documentation/git-filter-branch.txt
@@ -86,8 +86,7 @@ OPTIONS
 	This filter may be used if you only need to modify the environment
 	in which the commit will be performed.  Specifically, you might
 	want to rewrite the author/committer name/email/time environment
-	variables (see linkgit:git-commit-tree[1] for details).  Do not forget
-	to re-export the variables.
+	variables (see linkgit:git-commit-tree[1] for details).
 
 --tree-filter <command>::
 	This is the filter for rewriting the tree and its contents.
@@ -340,12 +339,10 @@ git filter-branch --env-filter '
 	if test "$GIT_AUTHOR_EMAIL" = "root@localhost"
 	then
 		GIT_AUTHOR_EMAIL=john@example.com
-		export GIT_AUTHOR_EMAIL
 	fi
 	if test "$GIT_COMMITTER_EMAIL" = "root@localhost"
 	then
 		GIT_COMMITTER_EMAIL=john@example.com
-		export GIT_COMMITTER_EMAIL
 	fi
 ' -- --all
 --------------------------------------------------------
-- 
2.13.0

Re: [PATCH] doc: filter-branch does not require re-export of vars

[Jeff King]

On Fri, May 26, 2017 at 07:36:54PM +0200, Andreas Heiduk wrote:

> The function `set_ident` in `filter-branch` exported the variables
> GIT_(AUTHOR|COMMITTER)_(NAME|EMAIL|DATE) at least since 6f6826c52b in 2007.
> Therefore the filter scripts don't need to re-eport them again.

Some old shells keep separate values for the internal and exporter
versions of variables. I.e., this:

  foo=one
  export foo
  foo=two

would continue to export $foo as "one", even though it is "two" inside
the script.

However, I think POSIX mandates the behavior you'd expect. And the only
shell I know that misbehaves in this way is Solaris /bin/sh, which we
have already declared too broken to support. According to

  https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/html_node/Limitations-of-Builtins.html#export

it sounds like there are some other antique shells which may do the
same (it doesn't cover this case explicitly, but the "coexist" cases it
mentions are likely to behave in this way).

At this point, I'd be inclined to remove the text as you suggest and
either make a small note at the bottom of the page, or just omit it
entirely and assume that anybody on an old non-POSIX shell can fend for
themselves.

-Peff

Re: [PATCH] doc: filter-branch does not require re-export of vars

[Junio C Hamano]

Jeff King <peff@peff.net> writes:

> On Fri, May 26, 2017 at 07:36:54PM +0200, Andreas Heiduk wrote:
>
>> The function `set_ident` in `filter-branch` exported the variables
>> GIT_(AUTHOR|COMMITTER)_(NAME|EMAIL|DATE) at least since 6f6826c52b in 2007.
>> Therefore the filter scripts don't need to re-eport them again.
>
> Some old shells keep separate values for the internal and exporter
> versions of variables. I.e., this:
>
>   foo=one
>   export foo
>   foo=two
>
> would continue to export $foo as "one", even though it is "two" inside
> the script.

Yup, that sounds like a grandfather's war story at this point,
though ;-)

> However, I think POSIX mandates the behavior you'd expect. ...
> ...
> At this point, I'd be inclined to remove the text as you suggest and
> either make a small note at the bottom of the page, or just omit it
> entirely and assume that anybody on an old non-POSIX shell can fend for
> themselves.

Sounds good to me.  Thanks.

Re: [PATCH] doc: filter-branch does not require re-export of vars

[Samuel Lijin]

On Fri, May 26, 2017 at 2:37 PM, Jeff King <peff@peff.net> wrote:
>
> On Fri, May 26, 2017 at 07:36:54PM +0200, Andreas Heiduk wrote:
>
> > The function `set_ident` in `filter-branch` exported the variables
> > GIT_(AUTHOR|COMMITTER)_(NAME|EMAIL|DATE) at least since 6f6826c52b in 2007.
> > Therefore the filter scripts don't need to re-eport them again.
>
> Some old shells keep separate values for the internal and exporter
> versions of variables. I.e., this:
>
>   foo=one
>   export foo
>   foo=two
>
> would continue to export $foo as "one", even though it is "two" inside
> the script.
>
> However, I think POSIX mandates the behavior you'd expect. And the only
> shell I know that misbehaves in this way is Solaris /bin/sh, which we
> have already declared too broken to support.

Off-topic, but where is this explicitly documented?

> According to
>
>   https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/html_node/Limitations-of-Builtins.html#export
>
> it sounds like there are some other antique shells which may do the
> same (it doesn't cover this case explicitly, but the "coexist" cases it
> mentions are likely to behave in this way).
>
> At this point, I'd be inclined to remove the text as you suggest and
> either make a small note at the bottom of the page, or just omit it
> entirely and assume that anybody on an old non-POSIX shell can fend for
> themselves.
>
> -Peff

Re: [PATCH] doc: filter-branch does not require re-export of vars

[Jeff King]

On Sun, May 28, 2017 at 09:35:30PM -0400, Samuel Lijin wrote:

> > However, I think POSIX mandates the behavior you'd expect. And the only
> > shell I know that misbehaves in this way is Solaris /bin/sh, which we
> > have already declared too broken to support.
> 
> Off-topic, but where is this explicitly documented?

I couldn't find a place that mentioned it explicitly, but POSIX defines
the concept as "the export attribute" of the variables. Which implies to
me that the bit is tied to the variable itself, not its value.

It also says that the flag is cleared when a variable is unset, so:

  foo=one
  export foo
  sh -c 'echo $foo should be one'
  unset foo
  foo=two
  sh -c 'echo $foo is not exported'

That could potentially affect somebody writing a filter-branch snippet,
but presumably if they are using "unset" they know what they're doing.

-Peff

Re: [PATCH] doc: filter-branch does not require re-export of vars

[Junio C Hamano]

Samuel Lijin <sxlijin@gmail.com> writes:

>> However, I think POSIX mandates the behavior you'd expect. And the only
>> shell I know that misbehaves in this way is Solaris /bin/sh, which we
>> have already declared too broken to support.
>
> Off-topic, but where is this explicitly documented?

One link I had readily available was

  https://public-inbox.org/git/7vei5qtnc5.fsf@alter.siamese.dyndns.org/

but there may be older discussions that were the actual process of
our officially having "written its /bin/sh off as broken and
unusable" if you dig further in the list archive.

Re: [PATCH] doc: filter-branch does not require re-export of vars

[Jeff King]

On Mon, May 29, 2017 at 01:27:45PM +0900, Junio C Hamano wrote:

> Samuel Lijin <sxlijin@gmail.com> writes:
> 
> >> However, I think POSIX mandates the behavior you'd expect. And the only
> >> shell I know that misbehaves in this way is Solaris /bin/sh, which we
> >> have already declared too broken to support.
> >
> > Off-topic, but where is this explicitly documented?
> 
> One link I had readily available was
> 
>   https://public-inbox.org/git/7vei5qtnc5.fsf@alter.siamese.dyndns.org/
> 
> but there may be older discussions that were the actual process of
> our officially having "written its /bin/sh off as broken and
> unusable" if you dig further in the list archive.

Ah, I had taken Samuel's question as "where is the POSIX behavior
documented". :)

The link above is a good explanation of the $() problem on Solaris. It
also doesn't do ${parameter#word}. I couldn't find a specific moment of
outlawing that /bin/sh, but there are several mentions of problems with
it going back to the beginning. Here's one from 2007:

  http://public-inbox.org/git/7vabt9sasl.fsf@assigned-by-dhcp.cox.net/

We had already given up on it by then.

I don't think there's anything in CodingGuidelines (though t/README does
call it "broken"), though it explicitly endorses $(), which rules out
Solaris /bin/sh.

I think we found in the last year or two that there are some old
versions of ksh that don't like our scripts (definitely ksh88, and maybe
ksh93?). Some light reading for the curious:

  http://public-inbox.org/git/CALR6jEhviK9KZxR6R6xzkZ5EAO-RjWj3xYah_DOSDXhEjYsT-A@mail.gmail.com/

  http://public-inbox.org/git/CALR6jEjWjJA0X2qXsxqObqc_yxrgX87LYf8cmJ0MmJFF6PkmTQ@mail.gmail.com/

  http://public-inbox.org/git/xmqqinxt3kwq.fsf@gitster.mtv.corp.google.com/

The problems are around backslash-escaping, signal exit codes, and
doubled parentheses. There are some patches, but I think we decided not
to pursue it (I couldn't find that exact decision, but it's referenced
later on).

-Peff