[RFC/PATCH] t0081-*.sh: Fix failure of the 'long read' tests

[RFC/PATCH] t0081-*.sh: Fix failure of the 'long read' tests

[Ramsay Jones]

In particular, the subshell feeding the input pipe, using the
shell function generate_tens_of_lines(), dies early due to a
broken pipe (SIGPIPE). The test-line-buffer command reads all
of the data it needs and closes the pipe, but the generate_\
tens_of_lines() function is still attempting to write to the
pipe, since it is in an infinite loop. This causes the later
"kill $!" to fail, since the process no longer exists.

The infinite loop is caused by the loop variable increment
expression, which looks something like this:

    : $((i = $i + 1)) || return

This expression does not actually increment the loop variable
at all, so $i remains at zero, leading to the infinite loop.

A simple fix, for some shells, would be to remove the '$'
from the '$i' within the arithmetic expansion. However, some
older shells don't support assignment as part of an arithmetic
expansion. Therefore, in order to fix the problem, we replace
the loop increment expression with the simpler:

    i=$(($i + 1))

which solves the problem, while also being more portable.

Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
---

Junio,

I only remembered last night that I have been skipping some tests
on Linux (oops!); this fixes one of the failures for me. One of
the reasons for me simply skipping this test, rather than fixing
it up earlier, was that it looked like (indeed *is*) another
example of a syntax error due to using an older dash shell.
(Also, I thought there was some talk of scrapping this test).

However, when I tried bash and an up-to-date dash, the problem
changed from a syntax error to an infinite loop! ;-P

The reason for the RFC is that I don't understand why (apparently)
I'm the only person seeing this failure; I suspect that a newer
version of bash than I have does not have this problem (ie the
original arithmetic expansion actually works). dunno :-(

Also, I considered changing the kill statement thus:

    -	kill $! &&
    +	if kill -0 $! 2>/dev/null
    +	then
    +		kill $!
    +	fi &&

but I'm not sure that actually helps, and was not necessary to
fix the failure for me.

[BTW, the other tests failing for me are t4034 tests #21 (diff 
driver 'bibtex'), #25 (diff driver 'html); I haven't had time to
investigate yet]

ATB,
Ramsay Jones

 t/t0081-line-buffer.sh |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/t/t0081-line-buffer.sh b/t/t0081-line-buffer.sh
index 5067d1e..9b79e3b 100755
--- a/t/t0081-line-buffer.sh
+++ b/t/t0081-line-buffer.sh
@@ -25,8 +25,7 @@ generate_tens_of_lines () {
 		do
 			echo "$line"
 		done &&
-		: $((i = $i + 1)) ||
-		return
+		i=$(($i + 1))
 	done
 }
 
-- 
1.7.4

Re: [RFC/PATCH] t0081-*.sh: Fix failure of the 'long read' tests

[Jonathan Nieder]

Ramsay Jones wrote:

> --- a/t/t0081-line-buffer.sh
> +++ b/t/t0081-line-buffer.sh
> @@ -25,8 +25,7 @@ generate_tens_of_lines () {
>  		do
>  			echo "$line"
>  		done &&
> -		: $((i = $i + 1)) ||
> -		return
> +		i=$(($i + 1))
>  	done

This test is a mess.  Could you try the patch from the tip of

	git://repo.or.cz/git/jrn.git svn-fe

which just gets rid of it instead?

Thanks for a reminder.

Re: [RFC/PATCH] t0081-*.sh: Fix failure of the 'long read' tests

[Jeff King]

On Tue, Apr 26, 2011 at 07:05:38PM +0100, Ramsay Jones wrote:

> The infinite loop is caused by the loop variable increment
> expression, which looks something like this:
> 
>     : $((i = $i + 1)) || return
>
> [...]
>
> The reason for the RFC is that I don't understand why (apparently)
> I'm the only person seeing this failure; I suspect that a newer
> version of bash than I have does not have this problem (ie the
> original arithmetic expansion actually works). dunno :-(

Yeah, that syntax is handled just fine by my bash and dash:

  $ cat >foo.sh <<'EOF'
  i=1
  : $((i = $i + 1))
  echo $i
  EOF

  $ bash foo.sh
  2
  $ bash --version | head -n 1
  GNU bash, version 4.1.5(1)-release (x86_64-pc-linux-gnu)

  $ dash foo.sh
  2

But I think your i=$(($i + 1)) is the right solution.

-Peff

Re: [RFC/PATCH] t0081-*.sh: Fix failure of the 'long read' tests

[Ramsay Jones]

Jonathan Nieder wrote:
> Ramsay Jones wrote:
> 
>> --- a/t/t0081-line-buffer.sh
>> +++ b/t/t0081-line-buffer.sh
>> @@ -25,8 +25,7 @@ generate_tens_of_lines () {
>>  		do
>>  			echo "$line"
>>  		done &&
>> -		: $((i = $i + 1)) ||
>> -		return
>> +		i=$(($i + 1))
>>  	done
> 
> This test is a mess.  Could you try the patch from the tip of
> 
> 	git://repo.or.cz/git/jrn.git svn-fe
> 
> which just gets rid of it instead?

Ah, so I didn't imagine the discussion to remove this test!

Yes, this works great for me. So, I will keep skipping the test
until this patch makes it into git.git.

Thanks!

ATB,
Ramsay Jones

Re: [RFC/PATCH] t0081-*.sh: Fix failure of the 'long read' tests

[Ramsay Jones]

Jeff King wrote:
> Yeah, that syntax is handled just fine by my bash and dash:
> 
>   $ cat >foo.sh <<'EOF'
>   i=1
>   : $((i = $i + 1))
>   echo $i
>   EOF
> 
>   $ bash foo.sh
>   2
>   $ bash --version | head -n 1
>   GNU bash, version 4.1.5(1)-release (x86_64-pc-linux-gnu)
> 
>   $ dash foo.sh
>   2

Er, ... yeah, it works for my bash and (up-to-date) dash too!
[not the installed dash, of course, for which it is a syntax error]

Ahem ... *blush*

> But I think your i=$(($i + 1)) is the right solution.

Yes, this fixes the problem and does not introduce a regression.

So, the patch is correct, but (apart from the last sentence) the
commit message is *absolute rubbish*. I won't bore you with the
details of my lunacy! :-P

However, I much prefer Jonathan's patch which removes this test
completely!

ATB,
Ramsay Jones

[PULL svn-fe/maint] t0081-*.sh: Fix failure of the 'long read' tests

[Jonathan Nieder]

Hi,

Ramsay Jones wrote:

> So, the patch is correct, but (apart from the last sentence) the
> commit message is *absolute rubbish*. I won't bore you with the
> details of my lunacy! :-P
>
> However, I much prefer Jonathan's patch which removes this test
> completely!

Junio, please pull

  git://repo.or.cz/git/jrn.git svn-fe-maint

to receive the following fix.  The patch first visited the list two
months ago[1] and was discussed again last month[2] and seems to have
been well liked both times (well, I know I like it).

The tests it removes

 - are missing an EXECKEEPSPID prerequisite on Windows
 - use a : $((i = i + 1)) construct which does not seem to be portable
   to old versions of dash
 - are pointless, an eyesore, and a pain to maintain

What was the author thinking?  Sorry to let this sit for so long.  

Jonathan Nieder (1):
      Revert "t0081 (line-buffer): add buffering tests"

 t/t0081-line-buffer.sh |  106 +-----------------------------------------------
 1 files changed, 2 insertions(+), 104 deletions(-)

[1] http://thread.gmane.org/gmane.comp.version-control.git/170307/focus=170365
[2] http://thread.gmane.org/gmane.comp.version-control.git/172120/focus=172523

Re: [PULL svn-fe/maint] t0081-*.sh: Fix failure of the 'long read' tests

[Junio C Hamano]

Jonathan Nieder <jrnieder@gmail.com> writes:

> Junio, please pull
>
>   git://repo.or.cz/git/jrn.git svn-fe-maint
>
> to receive the following fix.  The patch first visited the list two
> months ago[1] and was discussed again last month[2] and seems to have
> been well liked both times (well, I know I like it).
>
> The tests it removes
>
>  - are missing an EXECKEEPSPID prerequisite on Windows
>  - use a : $((i = i + 1)) construct which does not seem to be portable
>    to old versions of dash
>  - are pointless, an eyesore, and a pain to maintain

Done, and thanks.