Ilia Mirkin

I'd like to second @ptomato's suggestion. IEEE 754 rounding makes sense for floating point contexts (maybe?), but for base-10 decimals, I think the Java BigDecimal is the gold standard --...

The problem is that the hex data you're (now) expecting is a sequence of pairs of hex digits followed by space, followed by another pair. My guess is that people...

FWIW the failing test is "ERROR: test_valid (frontend.selenium_tests.test_upload.UploadSeleniumTestCase)" which fails with a timeout. I don't think that's a result of anything done in this commit.

Hm, actually I can reproduce that selenium failure locally. I'll figure out what's going on. The binary upload gets a 500 error.

Ugh, the error is caused by frontend/views.py:123 doing edid_object.save() which results in a 'database table is locked: frontend_edid" error. Clearly not caused by my change (reproduced same thing on master),...

FWIW, I'm aware of this general issue. I have a "fix", but it doesn't work (hence the quotes). I was previously unaware that even simple shaders like this could hit...

0ff4/0ff8 are SCREEN_SCISSOR_HORIZ/VERT: https://github.com/envytools/envytools/blob/master/rnndb/graph/gf100_3d.xml#L444 And this suggests that a negative X/Y offset is being passed in (-9/-10 or so). This is used in nvc0_clear_render_target and nvc0_clear_depth_stencil. Which means that nine...

We could obviously clip it in nouveau, but IMHO the state tracker should be where validation lives, with minimal error checking in the driver.

Do these issues come up with wine + nouveau (but without using gallium-nine)? I.e. the same way that you tested with the nvidia blob driver, but with nouveau. If they...

@siro20 if you make traces and point me at how to replay them (I have a nine-patched wine now), I could have a look to see what's up.