Opened 12 months ago

Last modified 12 months ago

#15663 new bug

T9675 inexplicably regressed in allocations due to text submodule bump

Reported by: bgamari Owned by: alpmestan
Priority: normal Milestone: 8.6.1
Component: Compiler Version: 8.4.3
Keywords: Cc:
Operating System: Unknown/Multiple Architecture: Unknown/Multiple
Type of failure: None/Unknown Test Case:
Blocked By: Blocking:
Related Tickets: Differential Rev(s):
Wiki Page:

Description (last modified by bgamari)

Harbormaster somehow started showing a significant (~20%) regression in T9675's allocations starting in 989dca6cbd93205a72f12a0921ba1216559a9e1e. Strangely, this is a bump of the text submodule. This can't be right; something is extremely fishy here.

Change History (8)

comment:1 Changed 12 months ago by bgamari

Description: modified (diff)

comment:2 Changed 12 months ago by Ben Gamari <ben@…>

In fd89bb44/ghc:

testsuite: Bump expected allocations of T9675

This inexplicably started with 989dca6cbd93, which appears to be a bump of the
`text` submodule. This is very fishy so I've opened #15663 to ensure we
investigate.

comment:3 Changed 12 months ago by bgamari

Owner: set to alpmestan

Do you think you could look into this, Alp?

comment:4 Changed 12 months ago by alpmestan

Sure thing, I'll look into this next week.

comment:5 Changed 12 months ago by alpmestan

On my machine, the test passes both at commit e40b3888d7b70bb5c95c377cac66b2839332673d (the one before the text submodule bump) and at commit 989dca6cbd93205a72f12a0921ba1216559a9e1e (the text submodule bump).

Moreover, the important stats look really, really similar:

before text bump after text bump
bytes allocated 67745356 678640112
max bytes used 20177520 18842304
average bytes used 8714779 8859872
max mem in use 77594624 73400320

and it just goes on, most of those stats align. Is that submodule bump the exact moment when the problem appeared? I'm going to try with an older commit and see if the stats look similar there too.

comment:6 Changed 12 months ago by alpmestan

With a (master) commit from early september (ed789516e201e4fad771e5588da47a62e53b42b8):

bytes allocated 677459608
max bytes used 18826416
average bytes used 8745272
max mem in use 73400320

The figures are again very, very close what I got with the previous two commits. I'm now going to try with the tip of the master branch right now.

comment:7 Changed 12 months ago by alpmestan

With the tip of master from an hour or two ago (d90946cea1357d3e99805c27dab1e811785a4088):

bytes allocated 678812960
max bytes used 18871352
average bytes used 8865483
max mem in use 73400320

Those numbers do almost all look a bit superior to the ones I got with the commit from early september, but certainly not by 20%.

What am I missing?

comment:8 Changed 12 months ago by bgamari

All I know is that this Harbormaster build is the first failing build. This is all very confusing.

Note: See TracTickets for help on using tickets.