Pike-devel December 2003

pike-devel@lists.lysator.liu.se

12 participants
53 discussions

"thread_create"
by Jaroslaw Sitek 09 Dec '03

09 Dec '03

Hello I would like to ask if there were any changes in implementing threads between version 7.2 and 7.4 (stable) or rather what exactly has changed? The code of my function (implemented in master): object _load_thread(string path, mixed ... args) { program prg = _compile(path); string objid; object obj; array(string|int) err; if (objs[path]) return objs[path]; if (!prg) return 0; err = catch { obj = prg(@args); thread_create(obj->run); }; if (err) { werror(err[0]); werror(describe_backtrace(err[1])); } ... works perfectly in version 7.2, it creates suitable number of threads when it is called on a given objects. But the same function called on the same objects in version 7.4 creates only one (first) thread and though the compiler doesn't return any error messages when the function is initialized, threads are not 'loaded'. To check if threads are created I simply: ps aux|grep pike There are no names of threads, just the process is visibly divided into as many threads as the number of objects called in individual call. Could anyone tell me what is the problem? My system: Linux, Debian, i686 Aron@Valhalla

1 0

program(x)
by Mirar ＠ Pike developers forum 08 Dec '03

08 Dec '03

It seems to do something (I got an illegal assigment for x=peer_program() if I changed the type), but PikeParser doesn't allow it.

1 1

program(x)
by Mirar ＠ Pike developers forum 08 Dec '03

08 Dec '03

Is the construct program(x) legal, or even useful? It seems to be allowed. I would like to constraint a variable to only be able contain programs compliant to a certain class.

1 0

call_callback
by Mirar ＠ Pike developers forum 08 Dec '03

08 Dec '03

And call_outs.

2 4

call_callback
by Mirar ＠ Pike developers forum 08 Dec '03

08 Dec '03

Is there any utility function in Pike that does the following, or an approximate? void call_callback(function f,mixed ... args) { mixed err; err=catch { f(@args); return; }; werror(master()->describe_backtrace(err)); } ...preferably with typechecking. I find that I would like to put a wrapper around my callbacks a lot of the time, to be certain that the code following still is run even if the user (of my library-ish routines) makes an error.

4 6

Problems with MIME.Message
by Marcus Comstedt (ACROSS) (Hail Ilpalazzo!) ＠ Pike (-) developers forum 07 Dec '03

07 Dec '03

>The first one is a typical problem I get with mails from Eudora. You can >find the mail at http://qo-op.com/break_mime.txt The problem with this one is that the boundary delimiter of the outer multipart occurs as a prefix of the boundary delimiter for the inner multipart. RFC2046 explicitly forbids this: As stated previously, each body part is preceded by a boundary delimiter line that contains the boundary delimiter. The boundary delimiter MUST NOT appear inside any of the encapsulated parts, on a line by itself or as the prefix of any line. This implies that it is crucial that the composing agent be able to choose and specify a unique boundary parameter value that does not contain the boundary parameter value of an enclosing multipart as a prefix. Complain to the makers of Eudora. >Next is the error log I get from the LMTP server included in Pike 7.5. >It shows that MIME.Message simply brokes really too much. I run the LMTP >server with Pike 7.2.526 in Caudium. The problem here is the Content-Type line: Content-Type: text/html;iso-8859-1 This is a syntax error. It's supposed to be Content-Type: text/html;charset=iso-8859-1 Again, it's the mail that is broken, not MIME.Message. / Marcus Comstedt (ACROSS) (Hail Ilpalazzo!) Previous text: >2003-12-05 16:20: >Subject: Problems with MIME.Message >-------------------------------------------------------------------- >Hi, > >Last time I mail the Pike devel list about problems with MIME, you ask >me for some examples. Here are some I get on a production server. >I contact your privately since some of the data you can find in these >emails are private. > >The first one is a typical problem I get with mails from Eudora. You can >find the mail at http://qo-op.com/break_mime.txt > >With Pike 7.2.526 it fails with: > >multipart message improperly terminated >/usr/local/pike/7.2.526/lib/modules/MIME.pmod:880: > create("Content-Type: >multipart/related;\n\ttype=\"text/html\";\n\tboundar > y=\"=====================_197842502==_.REL\"\n",0,0) > >MIME: Message() > >And with 7.4.31: > >boundary missing from multipart-body >/usr/local/pike/7.4.31/lib/modules/MIME.pmod/module.pmod:1080: > MIME.Message()->create("Content-Type: >multipart/related;\n\ttype=\"text/ht > >ml\";\n\tboundary=\"=====================_197842502==_.REL\"\n",0,0,0) >object_program(MIME): MIME->Message() >/usr/local/pike/7.4.31/lib/modules/MIME.pmod/module.pmod:1100: > MIME.Message()->__lambda_65685_4_line_1090("Content-Type: >multipart/relate > >d;\n\ttype=\"text/html\";\n\tboundary=\"=====================_197842502==_ > .REL\"\n") > > >Next is the error log I get from the LMTP server included in Pike 7.5. >It shows that MIME.Message simply brokes really too much. I run the LMTP >server with Pike 7.2.526 in Caudium. > >I've added the following debug at line 869 of MIME.pmod to help debugging: > >write("headers[\"content-transfer-encoding\"]=%s\n", >headers["content-transfer-encoding"]); >array(string) arr=tokenize(headers["content-transfer-encoding"]); >write("arr=%O\n", arr); > >The log file also contains all the contents of the mail. I've tried to >find out where the problem but I didn't succeed. > >You can find the log file at http://smtp.linkeo.org/error_log.gz > >-- David Gourdelier > > > > / Brevbäraren >

1 0

Pike benchmarks
by Martin Nilsson (saturator) ＠ Pike (-) developers forum 05 Dec '03

05 Dec '03

Some new benchmarks 7.2.550 vs 7.4.35 vs 7.5.13 using shootout tests: Pike7.2 Pike7.4 Pike7.5 1: 5.62 4.50 2.52 (ackermann) 2: 10.27 6.72 6.84 (ary3) 3: 8.55 6.62 6.42 (fibo) 4: 5.92 4.78 4.61 (hash) 5: 11.05 7.22 6.09 (heapsort) 6: 7.46 6.69 6.56 (lists) 7: 8.19 8.25 6.09 (methcall) 8: 54.21 5.91 6.47 (nestedloop) 9: 6.04 5.23 4.99 (objinst) 10: 7.29 4.79 5.21 (random) 11: 9.95 6.71 4.54 (sieve) 12: 3.13 2.33 2.50 (strcat) Scoring based on position in tests (should probably be updated to include difference in percent): 1: Pike7.5 32 2: Pike7.4 27 3: Pike7.2 13

2 9

Types
by Martin Nilsson (saturator) ＠ Pike (-) developers forum 05 Dec '03

05 Dec '03

As an experiment I changed the type of stringp to tOr(tFunc(tStr,tInt1),tFunc(tNot(tStr),tInt0)) That does however cause compilation warnings that I think are wrong: Getopt.pmod:109: Warning: Calling non-function value. Getopt.pmod:109: Warning: Type called: function(string : int(1..1)) | function(!string : int(0..0))

2 2

C Autodocs
by Bill Welliver 05 Dec '03

05 Dec '03

I've run into what appears to be a limitation in the C autodoc extractor. The problem is that I want to have documentation in a module for a non-root module, such as System.PAM. Currently, that's not possible, and it failes with the following error: DocParser error: "@module: expected module name, got \" System.PAM\"" Tools.AutoDoc.DocParser.AutoDocError(SourcePosition(File: pam.cmod, lines: 60..60), "DocParser", "@module: expected module name, got \" System.PAM\"") Is there another way to do this, or does it require a change to the extractors? What would it take? I'm not familiar enough with the extractors to know where to begin (it looks some real monsters live in the code). If it were possible, would i still be able to add extra docs for the Pike portion? I don't think it would be as useful to have a little line about the pike module inheriting ___PAM or such. Any thoughts? Bill

2 2

More benchmarks
by Martin Nilsson (saturator) ＠ Pike (-) developers forum 04 Dec '03

04 Dec '03

Ok, I guess we are closing in on making a 7.6 release. This is the current shootout benchmark status against some other similar languages. Java PHP Perl Pike Python Ruby TCL 1: 3.60 12.56 4.91 0.63 4.59 9.40 Failed (ackermann) 2: 1.68 11.91 4.76 1.40 3.06 8.61 15.19 (ary3) 3: 2.89 12.13 6.37 1.55 4.46 7.00 11.63 (fibo) 4: 7.94 5.98 5.72 4.69 3.83 18.67 7.75 (hash) 5: 1.48 8.62 3.47 1.09 2.48 6.30 8.33 (heapsort) 6: 5.46 Missing 3.19 2.70 4.73 2.44 27.44 (lists) 7: 2.17 Missing 5.90 1.83 4.76 3.38 Missing (methcall) 8: 7.27 Failed 17.91 0.91 13.95 35.87 34.39 (nestedloop) 9: 3.61 Missing 12.99 3.37 5.99 6.16 Missing (objinst) 10: 1.81 5.80 2.23 1.43 2.92 5.93 5.87 (random) 11: 6.57 Failed 12.19 4.52 9.16 18.39 50.59 (sieve) 12: 3.12 4.50 2.32 1.90 4.21 3.80 7.02 (strcat) I have altered the lists test to use arrays, as intended, and not strings (which lowers the time from 13.1 seconds to 2.7). The languages are gcj/gij 2.3 (the compilation time is included), PHP 4.3.4RC3, Perl 5.8.2, Pike 7.5.13, Python 2.3.2, Ruby 1.8.0 and TCL 8.3. In the year since last benchmark we can see some significant improvements in the Python performance, almost 50% in many tests (though it is still not possible to run ackermann(3,9) without running out of stack). Perl is also doing good with about 25% better since 5.8.0. From test 8 we can see that multipass strength reduction isn't implemented in any other language. On the whole it looks like Perl and Python is now only twice as slow as Pike, so we better watch out...

9 13

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

Pike-devel December 2003