Pike-devel April 2007

pike-devel@lists.lysator.liu.se

3 participants
15 discussions

Process.run
by Peter Bortas (nu med mindre Opera) ＠ Pike (-) developers forum 15 Jul '08

15 Jul '08

I just checked in a Process.run() in 7.7: http://pike.ida.liu.se/development/cvs/diff.xml?module=Pike&file=7.7/lib/mo… The rational for that is that I was once again reminded by a user this week that getting Process.create_process() or Process.Process right without some corner case blocking is hard. It's not obvious that if you bind pipes to booth stderr and stdout you have to take in to account that OS buffers are not limetless. Process.run() is a thin wrapper around Process.Process that collects stdout, stderr and exitcode and return them in a mapping. This could potentially result in huge strings, but will do what most users want and even already do when they call create_process() with both pipes bound. Something like this should be in 7.8, but I'm open to suggestions for other things that might be usefull to return in the mapping or for changing member names.

5 7

error: `LIBS' was not set in the previous run
by Marcus Comstedt (ACROSS) (Hail Ilpalazzo!) ＠ Pike (-) developers forum 31 Jul '07

31 Jul '07

Has anyone else seen this problem? When building current 7.7 from scratch, it bombs out during the configuration of CommonLog: configure: loading cache ../.././config.cache configure: error: `LIBS' was not set in the previous run configure: error: changes in the environment can compromise the build In config.cache there is the following: ac_cv_env_LIBS_set= ac_cv_env_LIBS_value= (these are the only ac_env/ac_cv_env entries). Maybe this is related to the line LIBS= in PIKE_RETAIN_VARIABLES somehow? Interestingly enough, the problem does not occur on my PS3, which also has ac_cv_env_LIBS_set= ac_cv_env_LIBS_value= in config.cache...

3 8

Attributed types
by Henrik Grubbstr�m (Lysator) ＠ Pike (-) developers forum 03 May '07

03 May '07

As you all probably know, I've been tinkering with the type system in Pike 7.7 for a while now, and I've now reached the point of implementing attributed types. The question once again is "which syntax do you prefer?": GCC-style: void __attribute(noreturn) error(string msg); string werror(string __attribute(sprintf_format) fmt, mixed __attribute(sprintf_args) args); or (analogous with other pike types): __attribute(noreturn, void) throw(mixed); string werror(__attribute(sprintf_format, string) fmt, __attribute(sprintf_args, mixed) args);

4 6

Re: Returning 0 vs throwing errors in Stdio
by Mirar ＠ Pike developers forum 28 Apr '07

28 Apr '07

The main reason for this is that Pike has a very small user base, and invalidating lots of old code might mean that it will never work again, and it will scare away lots of people from Pike. Especially if the change is between 7.6 and a future 7.8.

4 5

Returning 0 vs throwing errors in Stdio
by Peter Bortas ＠ Pike developers forum 28 Apr '07

28 Apr '07

I'd like someone to expand on the advantages of the following change in 7.7: > o Stdio > The functions read_file(), read_bytes(), write_file() and > append_file() now always throw errors upon errors, to allow easier > use as errno doesn't have to be checked. read_file() and > read_bytes() still returns 0 if the file does not exists. Specificly for read_file() on files you lack permissions for. > Stdio.read_file("/etc/shadow"); (1) Result: 0 vs > Stdio.read_file("/etc/shadow"); Failed to open "/etc/shadow": Permission denied /pkg/pike/intel-debian3_1/7.7.13/pike/7.7.13/lib/modules/Stdio.pmod/module.pmod:1820: Stdio->read_file("/etc/shadow",UNDEFINED,UNDEFINED) As I see it it only makes it harder to properly handle errors. If Pike had a well developed system for handling errors it would be an improvement, but now it's only going to lead to more and buggier code. Unless someone can come up with a good explanation* for why it should be changed it should be reverted. * Like for example "I've extended and documented the error classes. It's obvious and clear how to use them now!". That'd be nice.

7 16

New Pike release: 7.6.112
by Peter Bortas ＠ Pike developers forum 28 Apr '07

28 Apr '07

There we go. Next up: Fixing the Debian packages. I do not have network at home this week, so that might take until next week though. Please upload binaries to pelix. / Peter Bortas Previous text: >15429783 2007-04-25 18:55 /123 lines/ Peter Bortas >Recipients: Pike (-) importmöte för mailinglistan >Subject: New Pike release: 7.6.112 >-------------------------------------------------------------------- >The Pike developers are proud to present Pike 7.6.112! > >It's also a historical moment in Pike release history; the win32 >installer is released together with the source! (And not a month later >after much pain.) > >Source and binaries can be found at http://pike.ida.liu.se/download/ > >Binaries for architectures other than win32 will turn up as people >finish building them. > > >Changelog for your reading pleasure: > >Changes since Pike 7.6.86 >---------------------------------------------------------------------- > >Changes in Pike: > >o Updated timezone data files. This includes the recent US DST change. >o Fixed mktime to work better with out of bounds time fields. [bug 4326] >o Fixed support for big endian 64-bit. >o Added conversion from Gmp.mpz objects to Gmp.bignum when cast to int. >o Added support for variable length records to Standards.IIM. >o Added timeout to Image.PS. Now kills the ghostscript process if it > hasn't returned after 30s. >o Added GL 1.2 and 1.3 constants. >o Added possibility to set offset and stride when creating textures from > memory objects in GL. (Not currently usable on win32.) >o Fixed CMYK calculation in Image.Image. >o Added adjusted_cmyk to Image.Image and use it when decoding with > Image.PS. >o Now handles more eps images internally in Image.PS without spawning > ghostscript. >o Added Standards.IIM that handles IPTC Information Interchange > Model data in JPEG or Postscript images created with PhotoShop. >o Added support for little endian TIFFs. >o Added support for TIFF to Image.Dims. >o Changed HTTP.Query to close connections explicitly. >o Changed sslfile to avoid blocking IO in destroy() as that could lock > up the gc. You should explicitly close() SSL connections since the close > packet might block. >o Fixed a bug in SSL module that could stop communication while > keeping the connection open indefinitely. >o General cleanup of SSL module. >o Improved Parser.XML.Validating. >o More paranoid checks of image headers when using Image.AVS. >o Fixes to make describe_value work as usual in the free pass of the > gc, so that e.g. sprintf("%O",...) behave well when called from > destroy(). >o Fixed truncated field names in ODBC. >o Fixed masking of file positions larger than 2 GiB on 64-bit. >o Fixed low-level parsing of HTTP headers to not reorder the headers. >o Added set_charset and get_charset proxy functions to Sql.Sql. >o Changed Protocols.HTTP.Query async_close() to care about the value > returned by ponder_answer(). Fixes requests to broken servers that > don't send proper results. >o Fixed bad free_string() in exit_bytecode(). >o Fixed an off-by-one crash when drawing alpha boxes all the way to > the bottom of the image. >o Fixed a multiset bug that could cause incorrect multiset sizes. >o Fixed a multiset bug that caused ints to be called as sorting > function when no sorting function was set. >o Fixed hour formatting in Calendar.Time.format_smtp(). >o Fixed TTL parsing of SRV records in Protocols.DNS. >o Fixed fatal "Unknown identifier type" that was caused by attempts > to encode programs that had not been fixated. >o Fixed some events in Calendar.Events. >o Fixed a double free in Shuffler. >o Fixed an inverted error test in Mysql.set_charset. >o Fixed a compilation crash of the Cache module on Pikes wihtout > threads. >o Fixed optimization bug for objects with a value of false. >o Fixed -t (tracing) on win32. >o Added support to Mysql for the charsetnr column attribute and proper > recognition of binary fields in unicode decode mode by using charsetnr > instead of the binary flag (which might be set for nonbinary columns > too, e.g. "VARCHAR(17) BINARY"). > > Unfortunately this means that unicode decode mode is disabled if > Pike is compiled with a client lib without support for the charsetnr > attribute, i.e. libmysql prior to v4.1.0. If you want to enable it > for compatibility reasons, set the environment variable > PIKE_BROKEN_MYSQL_UNICODE_MODE when starting pike. > >Other: > >o Improved Valgrind support. >o Now checks the return code from win32 VirtualProtect. This breaks > assembler optimized builds on Windows. It never worked in the first > place, but used to fail silently, so this is better. This has been > fixed in 7.7 by using another heap manager, but it has not been > decided if it should be be backported. >o Adjusted array growth policy in array_insert to double the size > instead of growing with approximately one eighth of the size. This to > ensure O(1) time complexity and lessen fragmentation. (This affects > ADT.Sequence._insert_element and ADT.CircularList._insert_element.) >o Fixed support for UnixODBC. >o Made IA32 machine code compatible with Darwin IA32 ABI. Enabled > machine code for Darwin. >o Upgraded bundled GMP from 4.2 to 4.2.1 > >Build fixes: > >o Support GNU Java (GCJ) >o Did away with the use of the undocumented internal MS win32 CRT > functions __loctotime_t and __dtoxmode. >o Fixed compilation of Nettle and Mysql on win32. >o Fixed typo that caused the configure script to search for lots of > nonapplicable include dirs on windows. >o Backported new INT_TYPE_{ADD,NEG}_OVERFLOW implementation from 7.7. > The old one does not work with GCC 4.1. >o Fixed problem where mysql_character_set_name exists in mysql.h but > not in the lib for some old mysql clients. >o Fixed epoll support to not require /dev/epoll on Linux. That means > epoll will be enabled all mainstream Linux distributions running on a > 2.6 kernel. If for some reason you want to disable it use > --without-epoll. >o Added Objective C support to the module build system. >o Fixed libnettle bundle compilation on win32. >o Fixed compilation of ffmpeg. >o Fixed compilation problem on OpenSolaris by checking for broken > Postgres environment. > >(15429783) /Peter Bortas/------------------------------------------- >

5 12

Pike 7.6.112 beta
by Peter Bortas ＠ Pike developers forum 23 Apr '07

23 Apr '07

http://pike.ida.liu.se/pub/pike/beta/7.6.112/Pike-v7.6.112.tar.gz http://pike.ida.liu.se/pub/pike/beta/7.6.112/Pike-v7.6.112.2-win32.exe Changes since last beta: o The Postfix callback crash fix. o Fixed icon in Windows start menu...

2 5

Re: Need another GC policy
by H. William Welliver III 23 Apr '07

23 Apr '07

It's probably worth noting that the Java garbage collector does suffer from many of the same problems, even if there are no "scheduled" runs. The main problem here is that the gc runs in a "stop the world" fashion. I'm not sure if that can be avoided completely. Bill >> I'm not well-known in gc algorithms, but a garbage collector like >> Java does not require scheduled runs, so no stalls there. Why not >> pike?

2 1

GLUE
by Mirar ＠ Pike developers forum 18 Apr '07

18 Apr '07

Now I'm missing GLUE.Font, base/Font.pike:14:Index Font not present in module "GLUE". base/Font.pike:14:Illegal program identifier: 0. base/Font.pike:42:Index Font not present in module "GLUE". base/Font.pike:44:Index Font not present in module "GLUE". base/bumpy.pike:14:Error looking up 'Font' in module '.'. I didn't see GLUE change when I updated, and it's in there in module.pmod. What up?

1 0

7.7 continues
by Mirar ＠ Pike developers forum 18 Apr '07

18 Apr '07

Trying to build a Pike with --with-long-long-int and --with-double-precision: | /home/mirar/pike/build/linux-2.6.17-11-generic-x86_64/pike -DNOT_INSTALLED -DPRECOMPILED_SEARCH_MORE -m/home/mirar/pike/build/linux-2.6.17-11-generic-x86_64/master.pike /home/mirar/pike/src/post_modules/Shuffler/make_sources.pike /home/mirar/pike/src/post_modules/Shuffler sources.h sources_to_compile | -:1: Type mismatch for callback function `!: | -:1: Expected: function( : int(2147483647..2147483647)) | -:1: Got : function( : int(0..1)) | -:1: Program constant "Iterator" is NULL. | make[2]: *** [override] Segmentation fault (core dumped) I can't even begin to spot what's going wrong here. Should I assume Pike 7.7 is so severly broken it's no use trying to get it to work?

3 3

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 April 2007