Re: [coreutils] [PATCH] tests: avoid portability problem with dash vs. local v=$splittable_val

[Pádraig Brady]

On 06/11/10 17:08, Jim Meyering wrote:
> Hi Pádraig,
> 
> When running make check SHELL=dash, I got one failure.
> It was in misc/stat-birthtime, due to its use retry_delay_:
> 
>   + num_sleeps=16
>   + test 5 -le 5
>   + gawk -v n=16 -v s=.07 BEGIN { for (i=0;i<n;i++) t = s" "t; print t }
>   + local delay=.07 .07 .07 .07 .07 .07 .07 .07 .07 .07 .07 .07 .07 .07 .07 
> .07
>   local: 1: .07: bad variable name
> 
> 
>>From bda912213e8605121226ce2e07951f1b682ef595 Mon Sep 17 00:00:00 2001
> From: Jim Meyering <address@hidden>
> Date: Sat, 6 Nov 2010 18:04:45 +0100
> Subject: [PATCH] tests: avoid portability problem with dash vs. local 
> v=$splittable_val
> 
> * tests/init.cfg (retry_delay_): Put "local var" in a statement before
> "var=...", to avoid portability problems with dash, when the RHS value
> contains white space.
> ---
>  tests/init.cfg |    5 +++--
>  1 files changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/tests/init.cfg b/tests/init.cfg
> index 7786257..fa1f0e8 100644
> --- a/tests/init.cfg
> +++ b/tests/init.cfg
> @@ -390,8 +390,9 @@ retry_delay_()
>    local num_sleeps=$attempt
>    local time_fail
>    while test $attempt -le $max_n_tries; do
> -    local delay=$($AWK -v n=$num_sleeps -v s="$init_delay" \
> -                  'BEGIN { for (i=0;i<n;i++) t = s" "t; print t }')
> +    local delay
> +    delay=$($AWK -v n=$num_sleeps -v s="$init_delay" \
> +            'BEGIN { for (i=0;i<n;i++) t = s" "t; print t }')
>      "$test_func" "$delay" && { time_fail=0; break; } || time_fail=1
>      attempt=$(expr $attempt + 1)
>      num_sleeps=$(expr $num_sleeps '*' 2)
> --
> 1.7.3.2.193.g78bbb

I see that issue with my dash-0.5.5.1-2.fc11 also:

$ dash -c 's="1 1"; ok=$s; local l=$s'
local: 1: 1: bad variable name

BTW, the above should give another error because "local"
is used outside a function. It does when within a script
that's executed, but not when the script (file name)
is passed as a parameter to dash?

I also tested the following (generate a single float)
which seems a bit cleaner to me:

diff --git a/tests/init.cfg b/tests/init.cfg
index 1048cf3..d704f34 100644
--- a/tests/init.cfg
+++ b/tests/init.cfg
@@ -388,8 +388,8 @@ retry_delay_()
   local num_sleeps=$attempt
   local time_fail
   while test $attempt -le $max_n_tries; do
-    local delay=$($AWK -v n=$num_sleeps -v s="$init_delay" \
-                  'BEGIN { for (i=0;i<n;i++) t = s" "t; print t }')
+    local delay=$($AWK -v n=$num_sleeps -v s=$init_delay \
+                  'BEGIN { print s*n }')
     "$test_func" "$delay" && { time_fail=0; break; } || time_fail=1
     attempt=$(expr $attempt + 1)
     num_sleeps=$(expr $num_sleeps '*' 2)

I tested all retry_delay_ using tests with dash like:
(cd tests && make check TESTS=misc/stat-birthtime VERBOSE=yes SHELL=dash)
Note that required fixing the 'local saved_IFS=$IFS' in tests/init.cfg also.
Are we going to disallow dash because of this "local" issue, or are
we going to change all uses of local name=$val to local name; name=$val ?
Note by using the single float patch above you don't need to change
any of the local delay="$1" lines in the tests ($1 is quoted in any case).
In fact I can't see any uses of "local" under tests that need to change,
but we may want to do so for style/consistency reasons.

cheers,
Pádraig.