| commit | b9c34b9a26dfacab2add7734850d1dedb4762d41 | |
| author | Mark Williams <mwilliams@fb.com> | |
| Sun, 10 Mar 2019 19:13:58 +0000 (10 12:13 -0700) | ||
| committer | Hhvm Bot <hhvm-bot@users.noreply.github.com> | |
| Sun, 10 Mar 2019 19:17:49 +0000 (10 12:17 -0700) | ||
| tree | ea10005dbb3e54daa0fff701d33d9d7457f214e5 | treesnapshot (tar.gz zip) |
| parent | 9bb43e54fbe9cebf72e235e888b34382eb2d0c17 | commitdiff |
Use BlockIds in rpoBlocks
Summary:
When we switch to copy_ptr, mutating a block could modify its pointer,
so we don't want to store long lived pointers in the FuncAnalysis object.
Using block ids solves that problem, and also reduces the memory
requirements. It also turns out that a large number of users of
rpoBlocks only wanted the blockid anyway.
Reviewed By: ricklavoie
Differential Revision: D14335998
fbshipit-source-id: 67f22cf9139d2df42c8f32a808cb21b8d4efe9c9
Summary:
When we switch to copy_ptr, mutating a block could modify its pointer,
so we don't want to store long lived pointers in the FuncAnalysis object.
Using block ids solves that problem, and also reduces the memory
requirements. It also turns out that a large number of users of
rpoBlocks only wanted the blockid anyway.
Reviewed By: ricklavoie
Differential Revision: D14335998
fbshipit-source-id: 67f22cf9139d2df42c8f32a808cb21b8d4efe9c9