On Tue, 2008-11-18 at 12:15 +0100, Jiří Paleček wrote: > On Tue, 18 Nov 2008 08:31:55 +0100, Subrata Modak > <[EMAIL PROTECTED]> wrote: > > > > > On Tue, 2008-11-18 at 01:15 +0100, Jiří Paleček wrote: > >> Hello, > >> > >> On Mon, 17 Nov 2008 12:19:13 +0100, Subrata Modak > >> <[EMAIL PROTECTED]> wrote: > >> > >> > > >> > On Fri, 2008-11-14 at 12:40 -0500, Michael Kerrisk wrote: > >> >> > Here's the latest version, for review-n-test enjoyment: > >> >> > >> >> I made a few small tweaks to the changelog: wording fixes, and para > >> >> break fixes; other than that, changed the text to say that I rewrote > >> >> (rather than updated) the test program. Please substitute it with > >> the > >> >> following. > >> >> > >> > I just did a quick patch for your test to run on LTP. If you have no > >> > issue(s), can we add this to LTP ? > >> > > >> > LTP-list, > >> > > >> > Can you also enrich this patch ? > >> > >> I have some nitpicks, in decreasing severity: > >> > >> First, the syscall, I believe, is not targeted at i386 and x86-64 only. > >> Therefore, it is not wise to have these explicitly mentioned in the > >> code. > >> Also, it would be better not to "#error" if the arch isn't one of those > >> fortunate, because ltp should build on others too. This should be fixed > >> by > >> patch 1. > >> > >> Disclaimer: This patch should make it compile (and fail at runtime with > >> TCONF) on all kernels that don't have the syscall, and actually run the > >> test on all kernels that do, depending on kernel headers version. > >> However, > >> I didn't test this (especially the selection of the syscall), so it > >> needs > >> to be checked. > >> > >> Second, if any of the syscalls vital for the test fails, it's preferable > >> to output the error message too, and call tst_brk() for cleanup (patch > >> 2). > >> > >> Third, there it would probably be better to use TFAIL/TPASS for > >> recording > >> success and failure instead of manual boolean flags (patch 3). > >> > >> Last, I think a successful test should print as little as possible and > >> multiline messages like "calling syscall..." are really not that useful. > >> Patch 4 disables them. > > > Thanks Jiri for taking out time to investigate and coming out with a > > clean implementation for this new test case. I have the following > > points: > > > > 1) Patch no. 1 will fail to apply due to absence of > > ltp/testcases/kernel/include/stub-list in the original source code. I > > think you wanted to modify testcases/kernel/include/stub-list.orig. Even > > That's because vapier deleted that file 5 days ago without telling anyone > :( More on that in a separate mail. > > > without this file, the patch will fail to compile with the following > > error: > > > > cc -Wall -I../../include -g -Wall -I../../../../include -Wall > > accept4_01.c -L../../../../lib -lltp -o accept4_01 > > accept4_01.c: In function ‘accept4’: > > accept4_01.c:176: error: ‘__NR_accept4’ undeclared (first use in this > > function) > > accept4_01.c:176: error: (Each undeclared identifier is reported only > > once > > accept4_01.c:176: error: for each function it appears in.) > > make[4]: *** [accept4_01] Error 1 > > The corrected (though untested) patch is attached. > > > 2) Other Patches apply with lots of HUNKS. Please take a clean diff. > > Well, they apply cleanly if you apply them in sequence, and they apply > fine individually (except for 4, which needs 3). I think I cannot do much > more when the patches all deal with the same file.
Thanks very much. Everything seems clean now. I have added the same to LTP. Regards-- Subrata > > > 3) Being too ambitious, i would also request you to kindly introduce > > other features that we normally have for a system call test in LTP like: > > i) No. of loops to run, > > ii) Time, etc, > > albeit in your spare time. > > I don't promise anything - the biggest problem with implementing these > features is that I don't know what are they for and how do they work, so I > would probably just copy&paste them from a different test. > > > Regards > Jiri Palecek > -- > Using Opera's revolutionary e-mail client: http://www.opera.com/mail/ ------------------------------------------------------------------------- This SF.Net email is sponsored by the Moblin Your Move Developer's challenge Build the coolest Linux based applications with Moblin SDK & win great prizes Grand prize is a trip for two to an Open Source event anywhere in the world http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Ltp-list mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ltp-list
