Bug 189376

Summary: Regression test failed on network address related functions
Product: [Fedora] Fedora Reporter: Anton Guda <atu>
Component: gccAssignee: Jakub Jelinek <jakub>
Status: CLOSED UPSTREAM QA Contact:
Severity: medium Docs Contact:
Priority: medium    
Version: rawhideCC: tgl
Target Milestone: ---   
Target Release: ---   
Hardware: i586   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2006-04-24 15:48:14 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
regression diff file none

Description Anton Guda 2006-04-19 15:55:02 UTC 
Description of problem:
While (second) rebuilding on postgresql error
appears at the test stage: some network-mask
related functions returns bad results.
Previous rebuild with same rpm was fine (local build 31 mar 2006).

Version-Release number of selected component (if applicable):
postgresql-8.1.3-2
gcc-4.1.0-7
glibc-2.4-5

How reproducible:
now always


Steps to Reproduce:
1. rpmbuild -bi --target i586 postgresql.spec
  
Actual results:
build failed on test stage.

Expected results:
Build without errors.




Additional info:
Comment 1 Anton Guda 2006-04-19 15:55:02 UTC 
Created attachment 127989 [details]
regression diff file
Comment 2 Tom Lane 2006-04-19 16:43:11 UTC 
This looks to me like it could be a compiler bug, seeing that the same source
code passes this test fine on many other platforms.  Did you update gcc since
your last successful build?  Does the failure go away if you change the --target
spec?

The code in question is network_network() in src/backend/utils/adt/network.c
if you want to take a closer look.  I'm betting that gcc is misoptimizing the
shift-and-mask operations.
Comment 3 Anton Guda 2006-04-19 17:28:14 UTC 
Yes, gcc was updated every time as it was updated in ``development'' 
Fedora branch. Tested arch: i566, i686.
I will try to rewrite some code to find misoptimization.
And will update gcc once again.
Comment 4 Anton Guda 2006-04-22 11:18:05 UTC 
It's really a gcc bug.
The miscompilated code in src/backend/utils/adt/network.c: network_network:
  
  byte = 0;
  while (bits)
  {
   ...
  }
If bits == 25, the loop running 3 times, not 4,

If loop header rewrited as 

  while(bits>0)

all tests passed.

gcc-4.1.0-9
Comment 5 Tom Lane 2006-04-22 18:49:14 UTC 
Reclassifying as gcc bug.
Comment 6 Jakub Jelinek 2006-04-24 15:48:14 UTC 
Tracking thsi upstream.