| <<<Back 1 day (to 2018/10/25) | 20181026 |
fredross-perry | tor8, sebras, robin - take 2, that last commit was incomlete. Thanks. http://git.ghostscript.com/?p=user/fred/mupdf.git;a=commit;h=ea7b182700613aa481b30057fac82932bdfad8bc | 03:30.51 |
Robin_Watts | fredross-perry: lgtm | 08:17.59 |
sebras | fredross-perry: I noticed that DisplayList.search() also returns Rect[] according to DisplayList.java but Quad[] according to mupdf_native.c/mupdf_native.h would you mind adding this to the commit? | 09:11.04 |
| fredross-perry: other than that I'm happy with it. :) | 09:11.13 |
tor8 | Robin_Watts: three commits on mupdf tor/master for review, if you got some time over. | 13:30.49 |
| you can look at my bmpcmp for the diffs | 13:31.06 |
Robin_Watts | First 2 are fine. | 13:54.20 |
| Looking at the context one now. | 13:54.26 |
| ok, I see the logic in that one. | 13:58.38 |
| We could arguably hold ctx->error in a local pointer to avoid repeatedly dereferencing it in fz_push_try | 13:59.08 |
| cos C will have no way of knowing that it's safe cos of pointer aliasing. | 13:59.33 |
| Why haven't we lost fz_fake_throw ? | 14:00.11 |
| We have, I just can't read diffs. | 14:00.55 |
| OK. lgtm too. | 14:00.59 |
fredross-perry | Robin, sebras - will do stay tuned | 14:34.55 |
tor8 | Robin_Watts: going from ctx->error to ctx.error would possibly help dereferncing | 14:48.23 |
| Robin_Watts: but that's for later, I think | 14:48.45 |
| thanks for the review | 14:48.49 |
fredross-perry | third time's the charm. http://git.ghostscript.com/?p=user/fred/mupdf.git;a=commit;h=82faf2d5da25a509636689e62ea7765d0d5c5a8f | 15:58.26 |
| robin, sebras ^^ | 16:03.17 |
Robin_Watts | fredross-perry: lgtm | 16:04.02 |
fredross-perry | groovy thx. | 16:04.12 |
tor8 | fredross-perry: lgtm, and I even built it :) | 16:04.27 |
fredross-perry | lol | 16:04.39 |
| pushing... | 16:04.52 |
moolc | tor8: sadly sebras' semi-fix worked better for my case | 17:11.53 |
| ~ | 17:11.55 |
| - llpp ~/x/pic/vim2.png | 17:11.55 |
| warning: image too large for fixed point math: 519 x 284370 | 17:11.55 |
| | 17:11.58 |
tor8 | moolc: sebras semi-fix is wrong (it will overflow and draw the wrong thing) | 17:25.43 |
| moolc: change the #define PREC in draw-affine.c if 130k x 130k is too small for you | 17:26.16 |
| or just stop using ridiculously big images :P | 17:26.31 |
sebras | tor8: I'm not sure I fully understand why it will overflow? | 17:26.47 |
| tor8: that's why I was setting dolerp = 0, to prevent the overflow and not do lerping at all..? | 17:27.08 |
moolc | tor8: i can stop sure, i'm not the only mupdf user though :) besides.. i'm using tiled rendering so _should_ be immune, but alas | 17:28.04 |
sebras | ah, so that's the u,v coordinates and the stepping with fa, fb, fc, fd which cause the issue even if we don't lerp. | 17:33.54 |
| :qa | 17:34.03 |
| the only way to handle ridiculously large pictures would be to not use fixpoint for u,v,fa,fb,fc,fd. but yeah 128kx128k pictures are likely rare enough that it is not useful to bother about those. | 17:35.42 |
moolc | point being, it worked before it doesn't work now... and no changing PREC doesn't help :( | 17:37.14 |
sebras | moolc: so settinc PREC to 15 doesn't help? | 17:37.54 |
moolc | sebras: nope, http://ix.io/1q5Y | 17:42.18 |
tor8 | setting PREC to a *smaller* value could help. you'll need at largest 12 to support (32-12=20 bits) >= 250k^2 large images | 18:44.23 |
| images this big haven't worked ever, you've just not seen the errors since we used to never check the limits and overflow (and clamp) | 18:45.24 |
sebras | ehm.. I apparently shouldn't be touching code today.... :( | 18:45.37 |
moolc | tor8: thanks, setting it to 12 solved the vim2 issue.. | 18:47.59 |
| tor8: it "worked" for tiled rendering, that's how i found out... | 18:49.00 |
tor8 | tiled rendering might be pulling out a subset of the image (depending on the encoding, robin would know better) which isn't the full huge size | 18:51.36 |
| anyway, I'm off. have a good weekend! | 18:51.49 |
sebras | tor8: likewise! | 18:51.58 |
moolc | wish i was tor.. ircing from my own private train station... funky | 18:53.10 |
sebras | moolc: they're a pretty common ISP here. | 18:53.30 |
moolc | sebras: here? your back in Sverige? | 18:54.12 |
sebras | y | 18:54.17 |
moolc | æ
°å | 18:56.30 |
| Forward 1 day (to 2018/10/27)>>> | |