David Benjamin comments

Results 298 comments of


                                            David Benjamin

We shouldn't use atexit() in our libraries (or modules, engines and providers alike)

Looks like this has also come up in https://github.com/sfackler/rust-openssl/issues/1293 and https://github.com/openssl/openssl/issues/6214. `OPENSSL_INIT_NO_ATEXIT` was added, but that once again has an issue with the wrong defaults. Also given that individual libraries...

We shouldn't use atexit() in our libraries (or modules, engines and providers alike)

I think that is merely saying that two threads can't both call `exit`. TBH it's pretty horrifying that libc implementations don't guard against that one, but hopefully only the top-level...

Weird behavior when navigating away from request_ticket_v1.html

Ugh, I bet this is a quirk of when Chrome does and doesn't keep pages in the same process (unit of similar-origin related browsing contexts... `https://**.mit.edu` are all similar-origin). This...

Weird behavior when navigating away from request_ticket_v1.html

We may need to experiment a bit to see exactly what browsers do. The features argument isn't really all that standard between browsers. The spec officially says to just ignore...

Investigate compiling for Emscripten for Firefox

Emscripten can't do threads, which is a problem. But it looks like the threads in uqm are slowly going away? Grepping for `AssignTask` suggests they're only used for audio, and...

Investigate compiling for Emscripten for Firefox

This is not going to invert easily. Lots of nested "event loops" and stuff. (Any call to `DoInput`.) I think the best answer here may be some crazy additional layer...

Clicking the next video button on hulu sometimes hangs

Actually, the block seems to be that the viewer's requested an NPN_GetProperty as a response to the NPP_NewStream (probably Chromium-level interleaving) and never gets a response. Possibly some sort of...

Clicking the next video button on hulu sometimes hangs

Looking at hulu's javascript, it appears they make a synchronous ajax request after calling closeApp. It's possible this call is taking too long or (more likely) deadlocking. Should reproduce this...

Should UBSAN's float-divide-by-zero be a default?

On the Chromium side, we ended up disabling this in our UBSan builds. AIUI, all targets that use IEEE floating point define this operation. There are some niche targets that...

Should UBSAN's float-divide-by-zero be a default?

Since it sounds like the main thing keeping OSS-Fuzz on the incorrect behavior is inertia, I've gone ahead and uploaded https://github.com/google/oss-fuzz/pull/11567 to fix this. I think https://github.com/google/oss-fuzz/issues/10564#issuecomment-1614003494 underestimates the cost...