Pike-devel

pike-devel@lists.lysator.liu.se

3041 discussions

Mapping mixup
by Martin Nilsson (Opera Mini - AFK!) ＠ Pike (-) developers forum 06 Sep '11

06 Sep '11

We've seen this nasty little issue. The following code class C { mapping options = ([1:1]); void create() { werror("%O\n\n", options); } } void main() { C c = C(); c->options[3] = 3; C cc = C(); } will result in nilsson@riaa:~$ pike --dumpversion 7.9.5 nilsson@riaa:~$ uname -a Linux riaa 2.6.38-10-generic #46-Ubuntu SMP Tue Jun 28 15:07:17 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux nilsson@riaa:~$ pike test.pike ([ /* 1 element */ 1: 1 ]) ([ /* 2 elements */ 1: 1, 3: 3 ]) thought it works fine on nilsson@khora:~$ pike --dumpversion 7.9.5 nilsson@khora:~$ uname -a Linux khora 2.6.32-28-generic #55-Ubuntu SMP Mon Jan 10 21:21:01 UTC 2011 i686 GNU/Linux nilsson@khora:~$ pike test.pike ([ /* 1 element */ 1: 1 ]) ([ /* 1 element */ 1: 1 ])

3 12

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 23 Aug '11

23 Aug '11

On 04/12/2011 11:05 AM, Mirar @ Pike developers forum wrote: > That's an old bug that I thought would be fixed by now. Pike optimizes > the + call and calls `+ with more than one argument. The Calendar > module ignores all other arguments than the first. > > (I think it's that one anyway.) Ok, then I have to wait for the new debian package (still on 7.8.352)

8 12

7.8 segfault
by Bill Welliver 28 Jun '11

28 Jun '11

I've migrated a fairly complex application that was running fine under Pike 7.6 and am getting a number of different situations that cause segfaults. I'm not able to completely rule out the possibility that there's a non-core pike c-module that may be to blame, but the module code has been running well and without any problems of this sort for a number of years. The GDB backtraces look like this: Program received signal SIGSEGV, Segmentation fault. [Switching to Thread -145934672 (LWP 2582)] 0x081276e5 in low_object_index_no_free (to=0xffaaad58, o=0x88bfa78, f=67) at /root/Pike-v7.8.352/src/object.c:1252 1252 add_ref(to->u.dummy=dummy); (gdb) bt #0 0x081276e5 in low_object_index_no_free (to=0xffaaad58, o=0x88bfa78, f=67) at /root/Pike-v7.8.352/src/object.c:1252 #1 0x08133599 in index_no_free (to=0xffaaad58, what=0xf720f260, ind=0xffaaad60) at /root/Pike-v7.8.352/src/operators.c:87 #2 0x08076d57 in opcode_F_STRING_INDEX (arg1=702) at /home/zino/hack-local/Pike/7.8-distmaker/src/interpret_functions.h:1854 #3 0xf61c8a04 in ?? () #4 0x000002be in ?? () #5 0xf7610160 in __after_morecore_hook () from /lib/libc.so.6 #6 0xffaaad88 in ?? () #7 0x08a03870 in ?? () #8 0x00000000 in ?? () Any suggestions? This is 7.8.352 on an intel atom (64-bit debian). Best, Bill

2 1

possible sscanf misfeatures
by H. William Welliver III 22 Jun '11

22 Jun '11

I don't recall if anyone else has reported the following problems with sscanf() or not. I'm using 7.8.352 and having the following problems: 1. The + and - modifiers to %c don't seem to work in combination, even though it seems reasonable that one might want to decode a signed value in big endian order. Separately they work fine. 2. The - modifier does not work with %F. I'm currently using a workaround of reading in the bytes as an array, reversing them and then feeding them back into sscanf(). Bill

5 16

Re: possible sscanf misfeatures
by Mirar ＠ Pike developers forum 22 Jun '11

22 Jun '11

> sprintf("%+2c",17); (26) Result: "+\0\21" Can't ever be what you want...

1 0

foreign_ident frustration
by Martin Stjernholm, Roxen IS ＠ Pike developers forum 17 Apr '11

17 Apr '11

I'm getting quite fed up fiddling with those foreign_ident markers. A serious problem is that the changes in .gitattributes makes it cumbersome to cherry pick patches between branches and reorder commits (before push, of course). The .gitattributes diffs also preclude the possibility for git to recognize the same patch on the different branches. Another annoyance is that the commit checks get in the way: I never remember to check for foreign_idents before committing, so the commit get interrupted and I have to salvage the commit message (which in the case of using MaGIT in Emacs is definitely lost). Often it's also necessary to reset the file to get the $Id$ expansion right, which means I have to restage the commit with chunks and everything. On top of that there's various strange behavior in git with "spook differences" when the $Id$ handling changes. E.g. today I got into a state with a file where git reset/checkout resulted in it being considered changed in the $Id$ expansion right from the start. Killing the chunk didn't work, and it simply wasn't possible to get the working tree clean with the intended git tools. Things like this are bugs in git, but be that as it may, they add to the frustration. This foreign_ident system is very proper and correct, but it's frankly a big PITA, and I don't see any benefits with it that even comes close to make it worth the hassle. Keeping the cvs idents around in the historical versions is one thing, but replacing them gradually instead of wholesale is just a little more neat, perhaps. The whole thing boils down to avoiding just a single extra commit. I don't think it's a passing phase either - there are a lot of files, and it'll take a long time for them all to change naturally (it'll probably never happen). Also, the problem with backporting patches will only get worse as the current branch loses more foreign_idents compared to the old ones. So is there a compelling reason to keep those foreign_ident around, or would it be ok to drop them all in one go?

3 5

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 12 Apr '11

12 Apr '11

On 12-04-11 11:20, Mirar @ Pike developers forum wrote: > Or just do ->day(x)->minute(y) or so. :) Yes, but Calendar.Day()->day(-2)->minute(460); for example doesn't work. (At least in 7.8.352): > Calendar.Day()->day(-2)->minute(460); Not in range (Day 1..1 exist). /usr/lib/pike7.8/modules/Calendar.pmod/YMD.pike:818: Day(Tue 12 Apr 2011)->day(0) HilfeInput:1: HilfeInput()->___HilfeWrapper()

2 1

Calendar.Month / Calendar.Hour
by Marc Dirix 12 Apr '11

12 Apr '11

Hi, Pike 7.8.352 (Debian) In hilfe: > Calendar.now(); (21) Result: Fraction(Thu 7 Apr 2011 12:26:15.216691 CEST) > Calendar.Hour()+Calendar.Minute(); (22) Result: Minute(Thu 7 Apr 2011 12:01 CEST - 13:01 CEST) > Calendar.Month()+Calendar.Minute(); (23) Result: Minute(Fri 1 Apr 2011 0:01 CEST - Sun 1 May 2011 0:01 CEST) So Hour is calculated from the beginning of the last hour + 1 minute, and Month calculates starting of the month until next month + 1 minute. I expected the behaviour to be the same as with Calendar.Hour()? Best regards, Marc Dirix

4 6

Re: Calendar.Month / Calendar.Hour
by Marc Dirix 09 Apr '11

09 Apr '11

Thank you for thinking with me! I know the number of minutes, I just have to convert them to a date so I can test if current-time >= schedule-time. I think I need something along the lines of: next = Calendar.Day()->beginning() + Calendar.Minute()*x; or for tomorrow: next = Calendar.Day()->end() + Calendar.Minute()*x; /Marc

3 5

call_outs
by Arne Goedeke 07 Apr '11

07 Apr '11

I was about to change backend.cmod to use power of two hashtables, aswell. Its using hash_svalue anyhow, and it seems a little unnecessary to use mod prime hashtables there. Anyhow, while doing so and putting some benchmarks together I noticed that the code in backend.cmod:990 ff assumes that is_eq has side effets. Because of that it does alot of extra work. Its especially bad if one has alot of callouts to the same function since the buckets get very large. If a call to remove_call_out/call_out is the only thing that could change the hash tables, we might aswell do some kind of recursion detection and error instead (in case someone does call_outs from `==). Any thoughts? arne

4 11

Jump to page:

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