| | <<<Back 1 day (to 2018/04/26) | 20180427 |
| laertus | i'm getting this error: "mupdf: source/fitz/colorspace.c:1399: fast_rgb_to_cmyk: Assertion `"This should never happen" == NULL' failed." | 04:13.54 |
| | when trying to open this file: http://azarted.org/sites/default/files/BarguePlate9-12.pdf | 04:14.04 |
| | i'm using mupdf 1.12.0-r2 on amd64 gentoo linux | 04:14.34 |
| | also getting the same error when trying to open this file: http://www.davinciinitiative.org/uploads/3/8/6/4/3864305/drawing_with_envelopes.pdf | 04:17.09 |
| moolc | laertus: works here with git head | 04:19.35 |
| laertus | hmm | 04:19.50 |
| | confirmed | 04:49.58 |
| | git head works for me too | 04:50.02 |
| | i guess the problem's something to do with the gentoo version | 04:50.16 |
| | thanks for your help | 04:52.51 |
| avih | tor8: do you agree on the sort-is-stringifying-the array issue? | 08:22.05 |
| tor8 | avih: hang on, let me have a look | 08:33.40 |
| | avih: hmm, yes, I see the problem | 08:34.59 |
| | the toString call changes the stack slot as a nasty side effect | 08:35.15 |
| | so your patch looks correct | 08:35.25 |
| avih | exactly, which Sp_sort doesn't expect | 08:35.32 |
| | yeah | 08:35.35 |
| tor8 | it's an insert sort | 08:35.41 |
| avih | well, O(n^2) though. very slow very quickly | 08:36.08 |
| tor8 | much better than bubble, sometimes worse than quicksort | 08:36.09 |
| avih | i have a pure js iterative quicksort, and for 1000 items it's ~10x faster than the native sort | 08:37.31 |
| | for 10k items the native one is tens of seconds (i aborted), while the js quicksort one is 300ms | 08:38.33 |
| tor8 | avih: well, I wrote the insert sort because it's easy to implement and I needed to get it done quick. | 08:41.18 |
| avih | sure. and it works :) | 08:41.33 |
| tor8 | if you're up to coding up the quicksort in jsarray.c I'll gladly take it on | 08:41.35 |
| avih | that's my code https://pastebin.mozilla.org/9084042 | 08:41.46 |
| tor8 | avih: you can make your own function even faster if you avoid inner functions | 08:44.16 |
| | if mujs doesn't need to create a lexical scope with upvalues, then it can compile the functions to 'lightweight' functions which are faster | 08:44.44 |
| avih | correct, and also create create a stringified array first if not using an external comparator | 08:44.56 |
| | this was a proof of concept for comparing to the internal sort. it's not optimized. | 08:45.39 |
| | but i wanted to avoid polluting the global object | 08:46.13 |
| tor8 | avih: if you trigger a 'debugger' statement, the output will list the disassemble of the function which has a 'lightweight' line if the function is compiled to be lightweight | 08:46.17 |
| | avih: at least qsort_iterantive and partition are lightweigh | 08:47.18 |
| | qsort itself is not | 08:47.38 |
| avih | tor8: you mean "debugger" is a keyword which mujs understands? | 08:47.42 |
| tor8 | you could hoist the main body of work into a smaller inner function | 08:47.52 |
| | avih: yes. debugger is a standard javascript statement. | 08:48.02 |
| avih | yeah, it can be optimized a lot. also remove the call to the compare function if no external one is used. it's not optimized | 08:48.23 |
| | oh, didn't know about debugger. nice | 08:48.32 |
| tor8 | it prints the stack contents and a disassembly of all the functions in the call frame | 08:49.22 |
| | which also includes all inner functions of the functions and the script in the frame | 08:49.36 |
| avih | tor8: re lightweight vs non light function, where does the overhead manifest? while calling it? while resolving symbols? | 08:49.38 |
| tor8 | avih: both when calling it and when accessing local variables | 08:50.03 |
| avih | do you consider one or both general to js? or is it mujs specific? | 08:50.32 |
| tor8 | a heavyweight function needs an actual object to hold the local variables (since they can be accessed from the inner functions as upvalues, and need to be saved as part of the lexical scope if a function returns an inner function) | 08:50.50 |
| | so when calling a heavyweight function, we need to malloc and fill out an object for the stack frame instead of just using the simple stack | 08:51.15 |
| | and local variable accesses turn into dictionary lookups instead of indexing on the simple stack | 08:51.27 |
| | "getvar" vs "getlocal" opcodes | 08:51.37 |
| | this is mujs specific | 08:51.54 |
| avih | (sec, reading. admittedly, i haven't tried to understand when a function is/not lightweight) | 08:52.11 |
| tor8 | basically, if it has inner functions or uses the magic 'arguments' variable it becomes heavyweight | 08:52.46 |
| avih | in general though, i prefer not optimizing for mujs. | 08:52.49 |
| tor8 | the inner functions by themselves can be lightweight, it's just the parent function that has variables that need to be saved somewhere off the simple stack | 08:53.20 |
| | Math.floor involves looking up a property (Math) in the 'globals' object, then a property (floor) in the math object | 08:55.08 |
| | a faster trick if you want to turn something into an integer is ((right + left) / 2) | 0 | 08:55.25 |
| | do a bitwise or with zero | 08:55.30 |
| | that will force it to be a 32-bit integer for the bitwise math | 08:55.41 |
| | and then the bitwise math is a no-op | 08:55.55 |
| | that "trick" is what the asm.js javascript subset uses to annotate variables as being integers | 08:56.26 |
| avih | yeah, familiar with it. though by far the biggest factor in sort is the number of comparisons | 08:58.27 |
| moolc | shrugs | 08:59.03 |
| tor8 | avih: function cmpstr(a, b) { return (a = a+"") > (b = b+"") ? 1 : a == b ? 0 : -1; } might be a bit faster | 09:01.27 |
| avih | tor8: because it doesn't call a constructor from js code? | 09:01.59 |
| tor8 | it doesn't need to look up a global property and call it, exactly | 09:02.17 |
| avih | also, shouldn't it be ""+a rather than a+"" ? | 09:02.34 |
| tor8 | shouldn't matter, if either left or right side is a string, it becomes a string concatenation | 09:03.14 |
| avih | a+"" is not necessarily a string, while the spec says that they should be compared as strings | 09:03.20 |
| | but if a is 9 and b is 10 your code would say b is bigger, but the spec says b should be considered bigger here | 09:04.01 |
| | err.. the spec says a should be considered bigger | 09:04.25 |
| tor8 | 11.6.1 step 7 | 09:04.38 |
| | ""+a and a+"" should be equivalent | 09:05.11 |
| avih | huh, i thought the second arg is converted to the type of the first where possible | 09:05.42 |
| tor8 | nope. not that simple :) | 09:05.51 |
| avih | (just read 11.6.1-7) | 09:05.53 |
| tor8 | JS is full of nasty surprises | 09:06.51 |
| avih | at least there's a decent standard. unlike with posix shell :) | 09:08.02 |
| moolc | avih: HUH? | 09:08.10 |
| | avih: posix shell command language is perfectly defined | 09:08.21 |
| | it's the implemtnations that are interesting | 09:08.32 |
| avih | moolc: except where it leaves things undefined | 09:08.34 |
| moolc | avih: such as? | 09:08.39 |
| avih | A=foo mycommand | 09:08.52 |
| moolc | yes? | 09:09.04 |
| avih | does A remain foo after mycommand executes? | 09:09.05 |
| moolc | - A=foo echo "A=$A" | 09:09.56 |
| | A= | 09:09.57 |
| | | 09:09.57 |
| | it's not foo even during | 09:10.01 |
| avih | both depend on what "mycommand" is. whether it's a function or a builtin or a special builtin, and also on implementations | 09:10.45 |
| moolc | in zsh,bash,mksh,dash | 09:10.48 |
| avih | it would be foo if you replace echo with /bin/echo | 09:11.40 |
| moolc | avih: if you want A=foo; echo $a.. do it.. if you want (A=foo; echo $A) do that intead | 09:11.53 |
| | avih: no it wont | 09:12.10 |
| avih | moolc: i wasn't describing an issue i have. i was describing an issue the spec has | 09:12.18 |
| moolc | - A=foo /bin/echo "A=$A" | 09:12.19 |
| | A= | 09:12.19 |
| | | 09:12.19 |
| tor8 | avih: Array.prototype.sort fix on master now | 09:12.24 |
| avih | tor8: thx | 09:12.31 |
| moolc | avih: i'm not attacking you, i'm just having issues with the statement that posix doesn't defined shell command language properly | 09:13.07 |
| | -d | 09:13.26 |
| avih | moolc: right, sorry. of course it's not foo because "A=$A" is substituted while the line is parsed and before A is assigned | 09:13.45 |
| | i think. anyway, the posix shell command line spec is a hell compared to ecma262 | 09:14.31 |
| moolc | avih: it's rather vague and almost completely useless (once you want something that ought to be expressed by more than 250LOC) | 09:15.41 |
| avih | moolc: but my original question was whether A stays foo after the command. not during it | 09:15.52 |
| moolc | add a semicolon and it will | 09:16.28 |
| avih | you're solving the problem for me, a problem which i don't have. | 09:16.51 |
| moolc | avih: nope. i'm expanding my horizons. | 09:17.14 |
| | having recently converted some PSCL thing from PSCL to bash.. because - 'local' | 09:17.32 |
| avih | i'm just saying that even idiomatic expression like |A=foo mycommand| can behave differently depending on many factors | 09:17.47 |
| | moolc: you might be interested in this then https://github.com/avih/shcmp | 09:18.46 |
| moolc | avih: checking... you might be completely disinterested in following marvels - http://repo.or.cz/llpp.git/blob/master:/misc/llppac http://repo.or.cz/llpp.git/blob/master:/build.bash | 09:20.43 |
| avih | moolc: what am i looking at? | 09:22.21 |
| moolc | avih: a. build system in bash b. automatic anything to pdf converter | 09:22.58 |
| | example that shell can be useful | 09:23.20 |
| avih | funny. i wrote shcmp because i was writing a build system :) | 09:23.22 |
| | https://github.com/avih/mpv-build/tree/mpv-build-rewrite | 09:23.56 |
| moolc | avih: hah :) | 09:24.06 |
| | avih: i just switched to bash when lack of local became to painful | 09:24.18 |
| | s;to;too | 09:24.24 |
| avih | yeah, for now mine uses local. i'm considering removing it, but i probably won't. it's too useful | 09:24.46 |
| moolc | bash is too ubiquitous to ignore... | 09:25.14 |
| avih | but most shells support local. even yash which at my example shows it doesn't, support it with another command | 09:25.27 |
| | for me the shells which should work are dash/bash/busybox-ash and mksh (or bsd /bin/sh which is pretty mush the same) | 09:26.22 |
| moolc | avih: there are other quarter-pound like small differences that make pscl as opposed to specific implementation to big of pain | 09:26.25 |
| avih | mksh in general is quite good | 09:26.28 |
| moolc | avih: it is | 09:26.44 |
| | avih: you were/are trying to get rid of waf? | 09:27.07 |
| avih | no, mpv-build is a higher level. it builds few deps (mostly ffmpeg) which are typically not-latest in your average distro and links mpv statically with them | 09:28.04 |
| moolc | aha.. gotcha | 09:28.22 |
| | i've just expanded my build.sh to build local version of ocaml and build my stuff with that.. so yeah i can understand the pain | 09:28.51 |
| | avih: https://groups.google.com/forum/?_escaped_fragment_=msg/shake-build-system/2gQM0YDyDNs/DOKYnkWwCAAJ#!msg/shake-build-system/2gQM0YDyDNs/DOKYnkWwCAAJ | 09:29.41 |
| avih | moolc: it's actually fully generic and not specific to mpv. the main interface to configure what gets built is user.conf (which overlays default.conf - https://github.com/avih/mpv-build/blob/mpv-build-rewrite/config/default.conf ) | 09:30.03 |
| | the original mpv-build was mpv-specific, and there are still remains in my rewrite branch which are mpv-specific, but they're not part of the "core" system, like creating the debian package | 09:30.46 |
| moolc | avih: you are a peculiar man btw.. adding mujs to mpv (when it had/s lua) won't make you friends with tor here... then again mujs is his thing so maybe.. | 09:31.56 |
| avih | i wasn't looking to make friends with tor (though he does seem nice). i was looking to add js to mpv to make its scripting more accessible :) | 09:33.04 |
| sebras | tor8: the multi-page pnm/tiff-things LGTM. | 09:33.24 |
| moolc | avih: and place yourself into the world of ""+ vs +"" pain... okay... masochism is not yet punishable universally | 09:34.21 |
| sebras | tor8: I like "Add pdf_dcit_get_int etc." too, but I'd separate the implementation of those convenience functions from converting code to actually using them. | 09:34.51 |
| | because one commit then does something new, and the other is more or less a mechnical replacement. | 09:35.25 |
| avih | moolc: i tend to use constructs which i understand well, which is unrelated to the fact that admittedly i don't understand everything well. | 09:35.41 |
| | the ""+a/a+"" was from his code. in mine i use ""+a (if at all) and i 100% know how it behaves | 09:36.50 |
| | moolc: let me know if you find shcmp useful and/or has issues/suggestions with it | 09:39.44 |
| | have* | 09:39.50 |
| moolc | avih: likewise i'd appreciate the testing of repo.or.cz/llpp ;) | 09:41.09 |
| avih | sure, though i don't have use cases for it (i think? :) ) | 09:41.35 |
| moolc | avih: https://www.youtube.com/watch?v=qNszKpCUXhQ&list=PLLAukRknwSgFhpYsDKHY0oWpvV03Qj4AE ;) | 09:44.08 |
| | avih: just ran shells="sh dash bash mksh" ~/x/rcs/git/shcmp/shcmp ~/xsrc/llpp/misc/llppac ~/xsrc/cv/cven.html | 09:44.36 |
| | = sh, bash, bash_posix, busybox_ash, dash, mksh, mksh_posix | 09:44.36 |
| | no output | 09:44.40 |
| tor8 | sebras: easy enough to split | 09:44.51 |
| moolc | to be expected though.. i'm good | 09:44.54 |
| avih | lol | 09:45.28 |
| | moolc: should it print "<empty>" on such case? | 09:46.05 |
| moolc | the output contained an empty like... not "<empty>" verbatim | 09:46.38 |
| avih | i understand, which is ok. i'm asking if it should be enhanced to display "<empty>" if there's no output | 09:47.52 |
| | it currently prefixes the line with "<error> " if the command exited non-0 | 09:48.54 |
| | e.g. shcmp -c 'blahblah' | 09:49.59 |
| moolc | that violates one of the unix principles i suppose... but there's much debate lately on those | 09:50.30 |
| avih | yes, internally the clustering function is 100% pure, but it's called with arguments like prefix to print on error or success (such that you can use its output deterministically in your own code). so the wrapper uses empty prefix for success and <error> for error, and it can be enhanced easily to support an additional input to print if the output is empty. | 09:54.42 |
| | the output of the shcmp utility is not designed to be parsable. but the output of the internal cluster_outputs is | 09:56.23 |
| | so the core implementation doesn't violate principles i know of. but the utility does diverge from some to make it more useful. if you want, you can wrap the core cluster with your own code. | 09:59.04 |
| tor8 | sebras: updated commits on tor/master | 10:01.01 |
| moolc | avih: maybe some other time ;) | 10:01.13 |
| avih | it was a theoretical suggestion :) | 10:01.35 |
| sebras | tor8: "Add some comments." LGTM. | 11:09.02 |
| tor8 | avih: are you maybe confusing the ""+a and a+"" because of the silliness of {}+"" and ""+{} ? | 12:11.46 |
| | because those aren't what they look like, due to parser ambiguities | 12:13.29 |
| | the first one ({}+"") is not object plus string. it's empty code block statement, followed by unary positive on string) | 12:14.12 |
| | Forward 1 day (to 2018/04/28)>>> | |
Log of #mupdf at irc.freenode.net.