-
Notifications
You must be signed in to change notification settings - Fork 340
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Don't set a default maximum heapsize for Agda runs #7070
Comments
I'm in favour of removing the default limit. I think the argument was that people who need more than 3.5G for Agda are savvy enough to set their own RTS options, but I'm not sure I believe that setting the maximum heap size will make things faster... do we have an example that shows this? |
The plot thickens: this is the fourth project I've heard of that has recently (roughly since 2.6.4.1, although I could not establish a direct correlation with the update) started exceeding the maximum heap size. I mentioned this here before. Is this still a coincidence, or has something changed recently that made Agda more memory-hungry? |
@ncfavier wrote:
Good question! One could do a matrix-build (2.6.4 vs. 2.6.4.1; different heap limits) to find out if it is a regression in 2.6.4.1. |
For the project in question, there seems no obvious difference between 2.6.4 and 2.6.4.1: https://github.com/graded-type-theory/graded-type-theory/actions/runs/7708057745
2.6.4.1 is slightly faster (3%) than 2.6.4, but both either succeed or run out of memory in about 40 minutes. It seems like the heap increase was rather triggered by adding more code to the project: |
I experimented with different max heap sizes but they do not seem to make any impact (that has any significance) on the total running time for the SUT.
So I guess the conclusion is to remove the max heap pre-setting for Agda. |
Did you also try to omit the |
Started a test at: https://github.com/graded-type-theory/graded-type-theory/actions/runs/7742297113 Update: Blunder, if I do not set a limit there then the default limit |
I now ran a test with Agda nightly which doesn't have the default ✓ ci-matrix Matrix · 7743167201 JOBS |
This pre-setting makes Agda run out of heap when it could allocate more. Closes #7070.
This pre-setting makes Agda run out of heap when it could allocate more. Closes agda#7070.
The maximum heap size was removed in Agda 2.6.4.2 agda/agda#7070
The maximum heap size was removed in Agda 2.6.4.2 agda/agda#7070
This pre-setting makes Agda run out of heap when it could allocate more. Closes agda#7070.
A bit more than 4 years ago it was decided to set the maximum heap size for Agda to 3.5G:
agda/Agda.cabal
Line 864 in a9c1763
A CI run fails for me where this limit is reached: https://github.com/graded-type-theory/graded-type-theory/actions/runs/7666738532/job/20895151382
I was in vain searching workflow- and make-files to look for a setting of the maximum heap size, only to remember that I had seen something in
Agda.cabal
...From a UX perspective, I find it odd that we ship Agda with a default max heapsize.
Issue #3759 talks in the OP about a custom error should the heap be exhausted:
Such a nice error has never been implemented, so I suppose from a UX perspective we should also not set a max heap size by default.
ATTN (original contributors): @nad @WolframKahl @jespercockx @UlfNorell
The text was updated successfully, but these errors were encountered: