| <<<Back 1 day (to 2012/08/04) | 2012/08/05 |
sebras | morning ghostbot | 08:29.06 |
| hello! | 08:29.22 |
| hm.. I guess I'm in the ignore filter. :) | 08:29.35 |
Robin_Watts | Morning | 09:52.26 |
sebras | morning Robin_Watts | 10:49.21 |
| http://en.wikipedia.org/wiki/Comparison_of_OpenXPS_and_PDF it's a bit interesting that xps has been standardized by emca and pdf by iso, yet xps chose to use a 3d format standardized by iso and pdf one standardized by ecma. :) | 11:17.38 |
| I'm not 100% confident about the patch in pdf_image.c over at sebras/master. an alternative would be to move the FZ_INIT_STORABLE() and friends to before the colorspace is loaded (it's a bogus colorspace that triggers an exception and the image is not freed). | 13:17.50 |
rsc | Robin_Watts: compiling Ghostscript with "-march=armv6 -mfpu=vfp -mfloat-abi=softfp" instead of "-march=armv5tel -mfloat-abi=soft" reduced the time for a "gs -q -dNOPAUSE -sDEVICE=pdfwrite -sOutputFile=*.pdf *.ps -c quit" from ~ 15 to ~ 8 seconds | 15:28.53 |
Robin_Watts | Great. Does it work? :) | 15:29.08 |
rsc | Robin_Watts: well...I'm going to fire a rebuild with "-march=armv6z -mtune=arm1176jzf-s -mfpu=vfp -mfloat-abi=softfp" soon, that will hopefully bring it to ~ 5 seconds | 15:31.32 |
Robin_Watts | I still think you'd do better to move to a distro on which you don't have to use -mfloat-api=softfp. | 15:32.47 |
rsc | Robin_Watts: possibly. However an "openssl speed" with "-march=armv5tel -mfloat-abi=soft" already scales better than the standard Raspbian example for -mfloat-abi=softp | 15:34.18 |
| softfp | 15:34.38 |
| Robin_Watts: any pointer which other non-ghostscript internal library is heavily used when using pdfwrite? http://fpaste.org/EVO5/ | 15:36.33 |
Robin_Watts | rsc, sorry, no. | 15:50.24 |
rsc | Robin_Watts: nevertheless, thank you very much :) | 15:52.24 |
sebras | tor8: there is a freetype 2.4.10... | 16:31.39 |
| tor8: and I'm seeing a memory leak inside FT_Glyph_To_Bitmap() when it fails to open its resource. | 16:32.04 |
| tor8 Robin_Watts: in 0.9, if we encountered a content stream that consisted all of left brackets (256+ of them), we would parse all of them until we encountered EOF and the fail since we could not parse the innermost array. nowadays we fail earlier due to running out of error stacks for fz_try()... | 21:38.04 |
| so previously we failed with a proper error stack, but now we just exit() with "exception stack overflow". | 21:40.27 |
| the same goes for objects which have a large amount of recursive dicts. | 21:44.47 |
| tor8 Robin_Watts: 21 patches over at sebras/master to review. most under 15 lines, and couple >50 lines. | 23:10.46 |
| there are still a few patches from sumatrapdf that I ought to prepare for merging, but I'll wait until you guys have reviewed this batch. | 23:11.45 |
| goodnight ghostbot. | 23:18.18 |
Robin_Watts | goodnight, sebras :) | 23:18.36 |
| Forward 1 day (to 2012/08/06)>>> | |