Bug#750785: Slowness with context and xetex

Thanks Norbert. Was a pleasure working with you. Cheers, behdad On 14-08-19 10:53 AM, Norbert Preining wrote: > Hi Behdad, > > I just returned and saw with great pleasure a new release which > fixes the problem. > > I confirm that xetex now runs like hell again ;-) > > Thanks for the great s

Bug#750785: Slowness with context and xetex

Hi Behdad, I just returned and saw with great pleasure a new release which fixes the problem. I confirm that xetex now runs like hell again ;-) Thanks for the great support and help, I will surely disturb you again in case of problem. Thanks also to Hilmar for doing the last rounds of testing

Bug#750785: Slowness with context and xetex

On 12.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: Hi Behdad, > Attached. > I just wanted confirm that then patch solves the problem. The harfbuzz maintainer has uploaded a new package to Debian. Hilmar -- sigmentation fault signature.asc Description: Digital signature

Bug#750785: Slowness with context and xetex

On 14-08-13 12:14 PM, Hilmar Preusse wrote: > On 12.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: > > Hi Behdad, > >> Attached. >> > I just wanted confirm that then patch solves the problem. The > harfbuzz maintainer has uploaded a new package to Debian. Thanks Hilmer. I'll make a release to

Bug#750785: Slowness with context and xetex

-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Attached. On 14-08-12 05:31 PM, Hilmar Preusse wrote: > On 12.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: > > Hi Behdad, > >> https://github.com/behdad/harfbuzz/commit/8d5eebc0c6ada01128c6ee384340efdbef7ba29d >> > >> I'm failing to pull a raw

Bug#750785: Slowness with context and xetex

On 12.08.14 Norbert Preining (prein...@logic.at) wrote: Hi Norbert, > > This is a harfbuzz bug? Should be re-assign that bug? > > Let us wait. It could also be one of the many gcc 4.9 compiler bugs. > > Can you try compiling harfbuzz and/or xetex with gcc 4.8 instead > and see if that changes a

Bug#750785: Slowness with context and xetex

On 12.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: Hi Behdad, > https://github.com/behdad/harfbuzz/commit/8d5eebc0c6ada01128c6ee384340efdbef7ba29d > I'm failing to pull a raw unified diff out of github. Could you post it here? Tnanks, Hilmar -- sigmentation fault signature.asc Descript

Bug#750785: Slowness with context and xetex

The cache wasn't working if more than one (valid) shapers were passed in. Fixed now: https://github.com/behdad/harfbuzz/commit/8d5eebc0c6ada01128c6ee384340efdbef7ba29d behdad On 14-08-12 07:12 AM, Hilmar Preusse wrote: > On 11.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: > > Hi, > >> Befor

Bug#750785: Slowness with context and xetex

Ok, confirmed as HarfBuzz bug. Reproduced. Working on a fix. On 14-08-12 07:12 AM, Hilmar Preusse wrote: > On 11.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: > > Hi, > >> Before the return, add something like this: >> >> printf("DEBUG props %d features %d default_shaper %d shaper_list %d p

Bug#750785: Slowness with context and xetex

Humm. Ok, it looks suspicious, but also helpful. Let me think about it. On 14-08-12 07:12 AM, Hilmar Preusse wrote: > On 11.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: > > Hi, > >> Before the return, add something like this: >> >> printf("DEBUG props %d features %d default_shaper %d shape

Bug#750785: Slowness with context and xetex

On 11.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: Hi, > Before the return, add something like this: > > printf("DEBUG props %d features %d default_shaper %d shaper_list %d plan > func %p proposal func %p\n", > hb_segment_properties_equal (&shape_plan->props, &proposal->props), >

Bug#750785: Slowness with context and xetex

Hi Hilmar, > This is a harfbuzz bug? Should be re-assign that bug? Let us wait. It could also be one of the many gcc 4.9 compiler bugs. Can you try compiling harfbuzz and/or xetex with gcc 4.8 instead and see if that changes anything? Anyway, when I'm back from Hokkaido I can investigate furth

Bug#750785: Slowness with context and xetex

Thanks Hilmar. See below: On Mon, Aug 11, 2014 at 5:40 PM, Hilmar Preusse wrote: > > static hb_bool_t > hb_shape_plan_matches (const hb_shape_plan_t *shape_plan, >const hb_shape_plan_proposal_t *proposal) > { > return hb_segment_properties_equal (&shape_plan->p

Bug#750785: Slowness with context and xetex

On 11.08.14 Behdad Esfahbod (beh...@behdad.org) wrote: Hi Behdad, > That certainly isn't working as intended. I tested the cache locally (not in > the context of XeTeX) and it works fine. > At least I can reproduce the growing loops Norbert has seen, after applying the debug patch. > Can you d

Bug#750785: Slowness with context and xetex

Cold turkey! On 14-08-11 04:06 PM, Norbert Preining wrote: > I'll do, after I am back from holidays ;-) > > > PREINING, Norbert http://www.preining.info > > JAIST, Jap

Bug#750785: Slowness with context and xetex

I'll do, after I am back from holidays ;-) PREINING, Norbert http://www.preining.info JAIST, Japan TeX Live & Debian Developer GPG: 0x860CDC13 fp: F7D8 A928 26E3

Bug#750785: Slowness with context and xetex

That certainly isn't working as intended. I tested the cache locally (not in the context of XeTeX) and it works fine. Can you dig into hb_shape_plan_matches() and see why it's never matching? On 14-08-11 06:01 AM, Norbert Preining wrote: > Hi > > I think I found the bug: > > On Mon, 11 Aug 2

Bug#750785: Slowness with context and xetex

On 11.08.14 Norbert Preining (prein...@logic.at) wrote: Hi, > I think I found the bug: > Many thanks for investigation! > On Mon, 11 Aug 2014, Norbert Preining wrote: > > This is about 100 times faster total runtime on that document, with > > debug output, real 1.845s. > > > I recompile

Bug#750785: Slowness with context and xetex

Hi I think I found the bug: On Mon, 11 Aug 2014, Norbert Preining wrote: > This is about 100 times faster total runtime on that document, with > debug output, real 1.845s. I recompiled harfbuzz with some debug statements in the function, and it turned out that when I do that: (in src/hb

Bug#750785: Slowness with context and xetex

Hi, > This looks fine. Try passing NULL for shaperList and see if that changes > anything. This is about 100 times faster total runtime on that document, with debug output, real 1.845s. What can I try next? For completeness: Debian/sid, harfbuzz shared lib, amd64. Thanks a lot Norbert

Bug#750785: Slowness with context and xetex

This looks fine. Try passing NULL for shaperList and see if that changes anything. If it doesn't, then something in our shape_plan_cache isn't working. I have a hard time imagining what though. On Mon, Aug 11, 2014 at 1:55 AM, Norbert Preining wrote: > Hi Behdad, > > I have now done the foll

Bug#750785: Slowness with context and xetex

Hi Behdad, I have now done the following changes in XeTeXLayoutInterface.cpp (somewhere aboce the surrounding function, declar curtime and nxttime) static time_t curtime; static time_t nxttime; int layoutChars(XeTeXLayoutEngine engine, uint16_t chars[], int32_t offset, int32_t count, int32_t ma

Bug#750785: Slowness with context and xetex

Hi thanks > Ok, best that you get that printout first then. I will try as soon as I can, but I am leaving for long overdue holiday of one week tomorrow - without my laptop!!! ;-) :-D All the best Norbert PREINING, Norbe

Bug#750785: Slowness with context and xetex

On Sun, Aug 10, 2014 at 11:17 PM, Norbert Preining wrote: > Hi Behdad, > > thanks for your answer, and looking into it. > > On Sun, 10 Aug 2014, Behdad Esfahbod wrote: > > Not without more context (haha). > > I can give more context, easily, but I need some explanation > on *how* to print out inf

Bug#750785: Slowness with context and xetex

Hi Behdad, thanks for your answer, and looking into it. On Sun, 10 Aug 2014, Behdad Esfahbod wrote: > Not without more context (haha). I can give more context, easily, but I need some explanation on *how* to print out information concerning hbFace segment_props engine-fea

Bug#750785: Slowness with context and xetex

On 14-08-09 11:47 AM, Khaled Hosny wrote: > On Tue, Aug 05, 2014 at 03:04:03PM +0900, Norbert Preining wrote: >> Hi Khaled, >> >> concerning the slowness in xelatex and context/xetex, I found now >> which call is so delaying. I interspersed lots of printf and found that >> most time the xelatex pro

Bug#750785: Slowness with context and xetex

On Tue, Aug 05, 2014 at 03:04:03PM +0900, Norbert Preining wrote: > Hi Khaled, > > concerning the slowness in xelatex and context/xetex, I found now > which call is so delaying. I interspersed lots of printf and found that > most time the xelatex process is stucked in the following line > from XeT

Bug#750785: Slowness with context and xetex

Hi Khaled, concerning the slowness in xelatex and context/xetex, I found now which call is so delaying. I interspersed lots of printf and found that most time the xelatex process is stucked in the following line from XeTeXLayoutInterface.cpp: shape_plan = hb_shape_plan_create_cached(hbFace

Bug#750785: Slowness with context and xetex

On Mon, Jun 09, 2014 at 08:59:01PM +0900, Norbert Preining wrote: > Hi Khaled, > > (please keep Debian bug in Cc) > > I already have asked Taco and Hans, but it seems that the same problem > occurs also with xelatex, so I thought I ask you for ideas. > > We got a report here at Debian that conte

Bug#750785: Slowness with context and xetex

One more thing - I build the Debian package with -pg for profiling, but xetex runs did not produce a gmon.out, so there is something strange... On Mon, 09 Jun 2014, Norbert Preining wrote: > Hi Khaled, > > (please keep Debian bug in Cc) > > I already have asked Taco and Hans, but it seem

Bug#750785: Slowness with context and xetex

Hi Khaled, (please keep Debian bug in Cc) I already have asked Taco and Hans, but it seems that the same problem occurs also with xelatex, so I thought I ask you for ideas. We got a report here at Debian that context when run over xetex is extremely slow, which I can confirm. THe same happens wi

Bug#750785: Slowness with context and xetex

Hi Hans, thanks for your ideas. > Can you see if (in the background) the xetex font database gets > regenerated? I remember that long ago on windows we had a problem I didn't see anything running in the background. I also doubt that this is the reason, because xetex on linux will not call fc-cac

Bug#750785: Slowness with context and xetex

On 6/8/2014 9:26 AM, Norbert Preining wrote: Hi Hans, hi Taco, (please keep the Debian bug report in Cc) We got a report here at Debian that context when run over xetex is extremely slow, which I can confirm. It is interesting that when I drop the original TeX Live (not Debian) xetex into our

Bug#750785: Slowness with context and xetex

Hi Hans, hi Taco, (please keep the Debian bug report in Cc) We got a report here at Debian that context when run over xetex is extremely slow, which I can confirm. It is interesting that when I drop the original TeX Live (not Debian) xetex into our /usr/bin, then it is getting fast again. On th