Re: [PATCH v8 3/3] y2038: linux: Provide __clock_settime64 implementation

[Joseph Myers <joseph@codesourcery.com>]

On Wed, 18 Sep 2019, Lukasz Majewski wrote:

> Tests:
> - The code has been tested with x86_64/x86 (native compilation):
> make PARALLELMFLAGS="-j8" && make xcheck PARALLELMFLAGS="-j8"
> 
> - Run specific tests on ARM/x86 32bit systems (qemu):
> https://github.com/lmajewski/meta-y2038
> and run tests:
> https://github.com/lmajewski/y2038-tests/commits/master
> on kernels with and without 64 bit time support.

Could you give more details of the configurations (including kernel 
headers version, --enable-kernel version, kernel version used at runtime) 
for which you have built glibc and run the full glibc testsuite?  I 
suspect this code is pretty close to being ready to go in - but such 
patches have plenty of scope for mistakes (such as were in earlier RV32 
patch versions) such as typos, transposed parameters, etc., that are hard 
for humans to spot reliably and easy for compilers and testsuites to spot, 
so it's important to know that sufficient automated tests have been run to 
catch any such mistakes.

I gave a list in 
<https://sourceware.org/ml/libc-alpha/2019-08/msg00234.html> of five 
configurations I think are relevant to cover the different cases in 
patches such as this ("new" = 5.1 or later; Florian's changes to provide 
syscall tables in glibc would allow "#ifdef __NR_clock_settime64" to be 
removed and eliminate case (b)):

(a) one that has always had 64-bit time, e.g. x86_64;

(b) one with 32-bit time and old kernel headers (any kernel version at 
runtime);

(c) one with 32-bit time and new kernel headers, old kernel at runtime;

(d) one with 32-bit time and new kernel headers, new kernel at runtime but 
no --enable-kernel;

(e) one with 32-bit time and new kernel at runtime and new 
--enable-kernel.

If some of these are problematic to test, you can ask for help testing 
them.  For *this particular* patch you might not need to test both (c) and 
(d), because they are identical as far as compilation is concerned and the 
testsuite doesn't really cover execution of clock_settime anyway.  But 
it's in the nature of the Y2038 changes - involving lots of conditional 
code - that it's necessary to test in several different configurations to 
cover the conditionals adequately.

(For avoidance of doubt, it is *not* necessary to run build-many-glibcs.py 
for this patch, and nor would running build-many-glibcs.py be sufficient 
since it doesn't do execution testing or cover the kernel headers version 
and --enable-kernel variants.)

-- 
Joseph S. Myers
joseph@codesourcery.com