Set up an automated valgrind test #69

Activity

1. adamcrume

   added

   enhancement

   on Mar 10, 2017
4. Enet4 commented on Mar 14, 2017

   

   Enet4

   on Mar 14, 2017

   Contributor

   More actions

   Well, I read this recent blog post on using Valgrind for this very purpose, and we are still advised to change the memory allocation system (using the nightly alloc_system feature). Other than that, this is a useful resource to whoever is attempting to resolve this issue.
6. adamcrume commented on Mar 15, 2017

   

   adamcrume

   on Mar 15, 2017

   ContributorAuthor

   More actions

   We might copy the suggestion at rust-lang/rust#28224 (comment), which is to add a nightly feature and add this to the code:

   #![cfg_attr(feature="nightly", feature(alloc_system))]
   #[cfg(feature="nightly")]
   extern crate alloc_system;
   

   Then we can build on the nightly channel with --features=nightly and run valgrind against that.

   👍React with 👍1Reacted by Eduardo Pinho
8. adamcrume commented on Mar 21, 2017

   

   adamcrume

   on Mar 21, 2017

   ContributorAuthor

   More actions

   I'd like to see how a compiler-based sanitizer works when interacting with code it didn't compile, i.e. what happens when we allocate something through libtensorflow.so and then don't release that.

   I assume that the offsets in that stack trace could be traced back to a line of code via nm voodoo. Maybe they'll build that into the sanitizer at some point.
9. adamcrume

   added 2 commits that reference this issue on Jun 17, 2017

   Add run-valgrind

   ...

   22fd01c

   Add run-valgrind

   ...

   ddb684e
10. adamcrume

    added a commit that references this issue on Jul 29, 2017

    Add run-valgrind

    ...

    030860a
11. ramon-garcia

    added a commit that references this issue on May 20, 2023

    Parq wide - Implement horizontal (tensorflow#71)

    ...

    3fdeb0f