Michael Coblenz
@mcoblenz.bsky.social
Assistant Prof. at UC San Diego. Helping people write better software.
We rarely give Shakespeare credit for inventing alpha-conversion. “A rose by any other name…”
September 3, 2025 at 7:03 PM
We rarely give Shakespeare credit for inventing alpha-conversion. “A rose by any other name…”
Sounds like you already have two trained classifiers!
August 4, 2025 at 6:46 AM
Sounds like you already have two trained classifiers!
But maybe also I pick a different Mastodon server.
July 9, 2025 at 3:54 AM
But maybe also I pick a different Mastodon server.
Mastodon is too fragmented. I just focus on Bsky now.
July 9, 2025 at 3:53 AM
Mastodon is too fragmented. I just focus on Bsky now.
I can understand that. But I stopped checking Twitter long ago, so for me, this is it.
June 29, 2025 at 9:33 PM
I can understand that. But I stopped checking Twitter long ago, so for me, this is it.
Preprint of the paper, which my student Savitha Ravi will present: arxiv.org/pdf/2503.12309
arxiv.org
May 1, 2025 at 1:50 PM
Preprint of the paper, which my student Savitha Ravi will present: arxiv.org/pdf/2503.12309
Lots of great possibilities! Probably worth doing a study to see how various approaches would affect behavior.
April 11, 2025 at 6:28 AM
Lots of great possibilities! Probably worth doing a study to see how various approaches would affect behavior.
What will you do to make the environmental costs of using this transparent?
April 11, 2025 at 5:01 AM
What will you do to make the environmental costs of using this transparent?
Test passed! Awesome!
March 22, 2025 at 1:24 AM
Test passed! Awesome!
Geaxhurccswzvy4vi2ftz;
March 22, 2025 at 1:23 AM
Geaxhurccswzvy4vi2ftz;
You are right that the only way to be sure is to re-run the experiment.
March 12, 2025 at 4:54 PM
You are right that the only way to be sure is to re-run the experiment.
The API delta between the fixed version and the one we used in the experiment is rather trivial, so we felt the results likely generalized. But of course re-running the experiment would be best.
March 12, 2025 at 4:38 PM
The API delta between the fixed version and the one we used in the experiment is rather trivial, so we felt the results likely generalized. But of course re-running the experiment would be best.
Also as I understand it, the stack maps approach cannot be made thread safe, which is why I haven't felt motivated to finish this implementation; it still would not be of practical use. Really the whole GC implementation needs a redo.
March 12, 2025 at 3:54 PM
Also as I understand it, the stack maps approach cannot be made thread safe, which is why I haven't felt motivated to finish this implementation; it still would not be of practical use. Really the whole GC implementation needs a redo.
On the other hand, the main branch's version has GC enabled. The problem with this is that the mark implementation only handles certain cases, so this is a much more serious unsoundness problem (which is also documented in the paper and the code).
March 12, 2025 at 3:54 PM
On the other hand, the main branch's version has GC enabled. The problem with this is that the mark implementation only handles certain cases, so this is a much more serious unsoundness problem (which is also documented in the paper and the code).
I think this is the relevant commit. Note it’s on a branch: I never got around to merging. Criticism is welcome. github.com/mcoblenz/Bro...
Made as_ref and as_mut private. Removed Deref from GcRef. · mcoblenz/Bronze@34c1538
github.com
March 12, 2025 at 12:26 AM
I think this is the relevant commit. Note it’s on a branch: I never got around to merging. Criticism is welcome. github.com/mcoblenz/Bro...
Sorry I should have read the whole thread before replying! Still getting used to the Bluesky UI.
March 12, 2025 at 12:20 AM
Sorry I should have read the whole thread before replying! Still getting used to the Bluesky UI.
The implementation would definitely need a from-scratch redo for real work.
March 12, 2025 at 12:19 AM
The implementation would definitely need a from-scratch redo for real work.
Can you be more specific? We did fix some soundness problems post-experiment. And the implementation is a partial proof of concept and not a working GC.
March 12, 2025 at 12:17 AM
Can you be more specific? We did fix some soundness problems post-experiment. And the implementation is a partial proof of concept and not a working GC.
They’re pretty similar, right? Both are about avoiding side effects.
February 26, 2025 at 3:19 PM
They’re pretty similar, right? Both are about avoiding side effects.