google-perftools Google Group

Re: Issue 196 in google-perftools: valgrind errors for programs compiled with profiling

google-perfto...@googlecode.com — Sat, 28 Nov 2009 04:09:49 UT

Comment #1 on issue 196 by dstahlke: valgrind errors for programs compiled
with profiling
[link]

The full valgrind output...

Attachments:
vg.txt 28.8 KB

--
You received this message because you are listed in the owner
or CC fields of this issue, or because you starred this issue.

Issue 196 in google-perftools: valgrind errors for programs compiled with profiling

google-perfto...@googlecode.com — Sat, 28 Nov 2009 04:05:48 UT

Status: New
Owner: ----

New issue 196 by dstahlke: valgrind errors for programs compiled with
profiling
[link]

What steps will reproduce the problem?
1. echo 'int main() { }' > prof-vg-test.c
2. gcc -o prof-vg-test prof-vg-test.c -lprofiler -g

Re: google-perftools

mrab...@gmail.com (Mark Rabkin) — Fri, 27 Nov 2009 16:12:37 UT

The linker is probably throwing away your profiling symbols because it
thinks they're unused. Try either providing an ld option that tells is to
take the whole .a (not just the used parts), or try calling any profiling
function from your program, like this:

int main() {
ProfilerRegisterThread();

google-perftools

matthew.horsn...@gmail.com (Matt) — Fri, 27 Nov 2009 14:06:51 UT

Am I doing something wrong with the cpu performance tool. I build a
link against the .a files, and then simply set the CPUPROFILE
environment variable but no files are generated?

The process is a not a setuid process, anyone with any experience out
there that can help me out?

Re: Issue 195 in google-perftools: pprof doesn't use correct URI in http request while profiling

google-perfto...@googlecode.com — Wed, 25 Nov 2009 17:47:49 UT

Comment #4 on issue 195 by csilvers: pprof doesn't use correct URI in http
request while profiling
[link]

Right, you'll have to try it against svn trunk.

Or else, search for the equivalent place in the code -- it should be near
line 2450,

Re: Issue 194 in google-perftools: large environments causes abort

google-perfto...@googlecode.com — Wed, 25 Nov 2009 17:40:44 UT

Updates:
Status: Started

Comment #3 on issue 194 by csilvers: large environments causes abort
[link]

That's genius! I'll do it for the next release.

--
You received this message because you are listed in the owner
or CC fields of this issue, or because you starred this issue.

Re: Issue 195 in google-perftools: pprof doesn't use correct URI in http request while profiling

google-perfto...@googlecode.com — Wed, 25 Nov 2009 17:01:59 UT

Comment #3 on issue 195 by mohit.aron: pprof doesn't use correct URI in
http request while profiling
[link]

The problem is not just with CPU profiles - even heap profiles are broken.
That is,
using a URL [link] results in /pprof/heap to

Re: Issue 194 in google-perftools: large environments causes abort

google-perfto...@googlecode.com — Wed, 25 Nov 2009 15:59:14 UT

Comment #2 on issue 194 by dgscreamer: large environments causes abort
[link]

Why not change the code section to something like the following?
[avoiding both getenv and /proc/self/environ]

if (*envbuf == '\0') { // haven't read the environ yet

Re: Issue 193 in google-perftools: Implement malloc_usable_size (PATCH INCLUDED)

google-perfto...@googlecode.com — Wed, 25 Nov 2009 00:30:54 UT

Comment #3 on issue 193 by csilvers: Implement malloc_usable_size (PATCH
INCLUDED)
[link]

Right, which is why we provide it, though under a different name.

I'll take it under consideration to override it like we do malloc()/etc,
but my guess

Re: Issue 193 in google-perftools: Implement malloc_usable_size (PATCH INCLUDED)

google-perfto...@googlecode.com — Wed, 25 Nov 2009 00:12:36 UT

Comment #2 on issue 193 by ice799: Implement malloc_usable_size (PATCH
INCLUDED)
[link]

malloc_usable_size exists on FreeBSD and Linux. From what I can tell, it
does not
exist on Solaris.

It is pretty useful, though.

--
You received this message because you are listed in the owner

Re: Issue 194 in google-perftools: large environments causes abort

codesite-nore...@google.com — Tue, 24 Nov 2009 23:47:24 UT

Updates:
Labels: Type-Defect Priority-Medium

Comment #1 on issue 194 by csilvers: large environments causes abort
[link]

Yes, this is a known problem with reading from the environment -- we should
probably
document it somewhere. We can't use getenv() in this part of the codebase,

Re: Issue 193 in google-perftools: Implement malloc_usable_size (PATCH INCLUDED)

codesite-nore...@google.com — Tue, 24 Nov 2009 23:43:22 UT

Updates:
Status: WontFix
Labels: Type-Enhancement Priority-Medium

Comment #1 on issue 193 by csilvers: Implement malloc_usable_size (PATCH
INCLUDED)
[link]

We have equivalent functionality in MallocExtension::GetAllocatedS ize().

Re: Issue 195 in google-perftools: pprof doesn't use correct URI in http request while profiling

codesite-nore...@google.com — Tue, 24 Nov 2009 20:58:52 UT

Updates:
Status: Started

Comment #2 on issue 195 by csilvers: pprof doesn't use correct URI in http
request while profiling
[link]

Does this patch fix things for you?
---
@@ -2450,7 +2450,7 @@
my $wget_timeout;
if (($path =~ m/$PROFILE_PAGE/) || ($path =~ m/$PMUPROFILE_PAGE/)) {

Re: Issue 195 in google-perftools: pprof doesn't use correct URI in http request while profiling

codesite-nore...@google.com — Tue, 24 Nov 2009 20:54:49 UT

Updates:
Labels: Type-Defect Priority-Medium

Comment #1 on issue 195 by csilvers: pprof doesn't use correct URI in http
request while profiling
[link]

It looks like this broke in a recent change to support PMUPROFILE. Can you
confirm

Issue 195 in google-perftools: pprof doesn't use correct URI in http request while profiling

codesite-nore...@google.com — Tue, 24 Nov 2009 19:23:16 UT

Status: New
Owner: ----

New issue 195 by mohit.aron: pprof doesn't use correct URI in http request
while profiling
[link]

What steps will reproduce the problem?
1. Have pprof fetch profile data from a URL such as
[link]