Many pprof "address ??? is longer than address length 8" warnings with 32-bit binary on 64-bit system

[GoogleCodeExporter]

What steps will reproduce the problem?
1. Build a 32-bit perftools and 32-bit binary on a 64-bit CentOS 5 system
2. Run the binary with CPUPROFILE set
3. Run pprof on the output on a 64-bit system

What is the expected output? What do you see instead?

Many thousands of lines from pprof like:

Warning:  address 00000000082148c5 is longer than address length 8

What version of the product are you using? On what operating system?

google-perftools-1.8.3.  CentOS 5.x x86_64, but using 32-bit binaries.

Please provide any additional information below.

I'm using 32-bit binaries solely to avoid having to think about the 64-bit 
usage disclaimers :), I'd prefer to be using 64-bit.

It appears that the build system's "nm" (which is 64-bit) returns 64-bit 
addresses, even when the object in question is 32-bit.

A potential fix for this appears to be in HexExtend(), to add:
  $addr =~ s/^0*//;
right after the other sed.

Original issue reported on code.google.com by sarahfin...@gmail.com on 14 Oct 2011 at 8:01

Attachments:

• gpt.txt

[GoogleCodeExporter]

Makes sense to me.  I'll get a fix in for the next release.

Original comment by csilv...@gmail.com on 18 Oct 2011 at 5:44

• Changed state: Accepted
• Added labels: Priority-Medium, Type-Defect

[GoogleCodeExporter]

Okay, thanks from Julian H. O. Fortescue (who provided the fix).

Original comment by sarahfin...@gmail.com on 18 Oct 2011 at 3:48

[GoogleCodeExporter]

Give my regards to Julian!  That is one talented developer.

Original comment by csilv...@gmail.com on 19 Oct 2011 at 7:14

[GoogleCodeExporter]

Original comment by csilv...@gmail.com on 19 Oct 2011 at 10:19

• Changed state: Started

[GoogleCodeExporter]

This should be fixed in perftools 1.9, just released.

Original comment by csilv...@gmail.com on 23 Dec 2011 at 12:49

• Changed state: Fixed