Massive test failures

Marc Glisse marc.glisse at inria.fr
Tue Mar 24 00:23:05 CET 2026


On Mon, 23 Mar 2026, Torbjörn Granlund wrote:

> As you can see from https://gmplib.org/devel/tm/gmp/date the GMP
> autobuilds fail for lots of systems, but passes for others.
>
> Sample failure:
>
>    make[5]: Entering directory '/var/tmp/gmp-obj/bwl-stat-64/tests'
>    PASS: t-bswap
>    PASS: t-constants
>    PASS: t-count_zeros
>    PASS: t-hightomask
>    PASS: t-modlinv
>    PASS: t-popc
>    PASS: t-parity
>    PASS: t-sub
>    /bin/sh: line 12: test: +: integer expected
>    ============================================================================
>    Testsuite summary for GNU MP 6.3.0
>    ============================================================================
>    /bin/sh: line 29: test: -gt: unary operator expected
>    # TOTAL:
>    /bin/sh: line 29: test: -gt: unary operator expected
>    # PASS:
>    /bin/sh: line 29: test: -gt: unary operator expected
>
> So, what is on line 12 and 29 then?  Glad you asked.  But I have no clue
> as the script name is not printed.  The autohell Makefiles are totally
> unreadable, so trying to figure out what script is run is very far from
> straightforward straightforward.
>
> If anybody here has any idea, please let me know here!

I think line 12 is
         if test `expr $$fail + $$xpass + $$error` -eq 0; then \

and line 29 is
             if test $$maybe_colorize = yes && test $$count -gt 0; then \

but that doesn't explain why $$fail and others are suddenly empty on those 
platforms. If wc was missing, we would probably have a louder message.

> These failures seem to have commenced on the 15th of March.  That date,
> I committed a small change to the shell script which runs these tests:
>
>    *** .~/testclient.sh.~812~	Tue Mar  3 16:55:53 2026
>    --- testclient.sh	Sun Mar 15 13:44:55 2026
>    ***************
>    *** 9,14 ****
>    --- 9,15 ----
>      MAXTUNETIME=3600
>      SUPPRESS_tune=false
>      SUPPRESS_try=false
>    + SUPPRESS_try=true		# FIXME: CALLING_CONVENTIONS incompatible with modern C
>      SUPPRESS_demos=false
>      TUNEPROGS=allprogs
>      EXEEXT=

Maybe it doesn't like the inline comment? No, it is probably unrelated 
indeed.

-- 
Marc Glisse


More information about the gmp-devel mailing list