Builder libpcap-solaris10-amd64 Build #3184

Results:

Build successful

SourceStamp:

BuildSlave:

Reason:

The SingleBranchScheduler scheduler named 'schedule-libpcap-solaris10-amd64' triggered this build

Steps and Logfiles:

1. git update ( 6 secs )
   1. stdio
2. shell Set developer mode ( 0 secs )
   1. stdio
3. shell_1 './autogen.sh' ( 8 secs )
   1. stdio
4. shell_2 './configure' ( 1 mins, 14 secs )
   1. stdio
   2. config.log
5. shell_3 'gmake' ( 34 secs )
   1. stdio
6. shell_4 'gmake check' ( 9 mins, 0 secs )
   1. stdio
7. shell_5 'rm -rf ...' ( 0 secs )
   1. stdio
8. shell_6 'gmake install ...' ( 4 secs )
   1. stdio
9. trigger triggered trigger-tcpdump-solaris10-amd64 ( 8 mins, 48 secs )
   1. - no logs -
   2. tcpdump-solaris10-amd64 #5989

Build Properties:

Forced Build Properties:

Responsible Users:

1. Denis Ovsienko

   denisohnoyoudont@ovsienko.info

Timing:

All Changes:

:

1. Change #247049

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 29 Oct 2025 18:55:36
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	ebddc84406b07b71bcb2559b921b75baa20bc424

   Comments

   TESTrun: Refine a few assorted accept blocks.
   In MTP tests, for example, "sls (3)" tests a relevant code path, but
   "sls (3 or 3)" tests effects of the optimizer on a code path that is
   already covered with another test.  This causes unnecessary differences
   in the unoptimized program, so remove such aliases and expect the short
   program to be the unoptimized program.  Idem ATM tests.  For the length
   tests keep all the aliases, but split these into separate tests on
   similar grounds.  Use the 00 prefix to test octal for IPv4 and fix a
   comment.

   Changed files

   • testprogs/TESTrun

2. Change #247050

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 29 Oct 2025 18:55:36
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	fb47c0422f14514374290c4cd668fd748ec30901

   Comments

   TESTrun: Define "opt" and "unopt" in more accept blocks.

   Changed files

   • testprogs/TESTrun

3. Change #247051

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 29 Oct 2025 18:55:36
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	949f5fc9f6b6aeead97382565a004ebcae60a164

   Comments

   Get BPF_K and BPF_X right in gencode.c.
   Apparently, earlier I made an error in commit 87b9146, in that it is not
   right to replace five instances of a BPF_JMP|BPF_JEQ|BPF_X block with
   calls to gen_jmp(): gen_jmp() uses JMP() to produce a BPF_K branch
   instruction, which takes a constant for the "#k", but a BPF_X branch
   instruction does not take a constant.
   
   This error did not manifest in practice because BPF_K == 0x00 and BPF_X
   == 0x08, so passing BPF_X to gen_jmp() by coincidence worked as before.
   To make it obvious which variety of a branch block the code means to
   produce, replace gen_jmp() with gen_jmp_k() and gen_jmp_x(), each of
   which does its own correct thing.
   
   Whilst prototyping the above bug fix, in a few cases I had confused
   new_stmt() with gen_jmp() and omitted a mandatory value from the ORed
   bitmasks, which was not obvious until the tests failed.  To make such
   mistakes compile errors, have JMP() take two arguments and use the
   macro instead of composing the bitmasks in place.

   Changed files

   • gencode.c

4. Change #247052

   Category	libpcap
   Changed by	Denis Ovsienko <denisohnoyoudont@ovsienko.info>
   Changed at	Wed 29 Oct 2025 18:55:36
   Repository	https://git.tcpdump.org/libpcap
   Project	libpcap
   Branch	master
   Revision	81c00ffef3c70ee84df80a65df1f0856c90b9263

   Comments

   Do not special-case BPF_JEQ in gen_relation_internal().
   The function stores one operand in A and the other in X.  For BPF_JGT
   and BPF_JGE it generates "jgt x" and "jge x" respectively, which works
   exactly as required.  However, for BPF_JEQ it subtracts X from A, then
   compares A with 0.  Although the immediate effect of this is the same as
   that of "jeq x", the former requires an additional instruction.  Also it
   does not preserve the original value of the operand in A, which may get
   in the way of subsequent optimizations; indeed, optimizer.c:opt_peep()
   folds "sub x; jeq #0" into "jeq x".
   
   Thus in gen_relation_internal() use one gen_jmp_x() invocation for all
   cases.  In 13 accept blocks this eliminates from the unoptimized program
   one instruction per every instance of "EXPR1 == EXPR2" and
   "EXPR1 != EXPR2".

   Changed files

   • gencode.c
   • testprogs/TESTrun