Pike-devel

pike-devel@lists.lysator.liu.se

2 participants
3046 discussions

More aggressive gc of weak mappings in Pike 7.9
by Henrik Grubbstr�m (Lysator) ＠ Pike (-) developers forum 01 Jan '13

01 Jan '13

Hi. I've now implemented the new rehash behaviour that we discussed during the conference ~2 months ago. This means that weak mappings that contain the only reference to a reference-counted object will free the object when the hashtable is resized. Note that this only takes care of the most common case of weak mappings. The more complicated cases will still need the full gc. For reference: INDEX INDEX VALUE VALUE HANDLING REFTYPE CONTENT REFTYPE CONTENT ------------------------------------------ STRONG * STRONG * KEEP STRONG * WEAK UNREF KEEP STRONG * WEAK SINGLE FREE STRONG * WEAK MULTIPLE KEEP WEAK UNREF STRONG * KEEP WEAK SINGLE STRONG * FREE WEAK MULTIPLE STRONG * KEEP WEAK UNREF WEAK UNREF KEEP WEAK UNREF WEAK MULTIPLE KEEP WEAK SINGLE WEAK * FREE WEAK * WEAK SINGLE FREE WEAK MULTIPLE WEAK UNREF KEEP WEAK MULTIPLE WEAK MULTIPLE KEEP Legend: STRONG Strong (default) reference. WEAK Weak (set with set_weak_flag()) reference. * Any value. UNREF Value that is not reference counted (eg ints or floats). SINGLE Reference-counted value with a single reference. MULTIPLE Reference-counted value with more than one reference. KEEP Entry to be kept after the rehash. FREE Entry to be freed during the rehash. I would appreciate testing, and any comments about the behaviour (intended or otherwise). Happy New Year! /grubba

1 0

Warnings for unused variables
by Jonas Walld�n ＠ Pike developers forum 31 Dec '12

31 Dec '12

When compiling Pike with clang I get massive amounts of warnings for unused variables. This detracts quite a bit from finding more serious errors and I've got a patch ready to remove them. However, before pushing that I'd like to hear if people have objections to changing declarations like in these examples: - static void exit_memory(struct object *o) + static void exit_memory(struct object *UNUSED(o)) and - static int got_udp_event (struct fd_callback_box *box, int event) + static int got_udp_event (struct fd_callback_box *box, int DEBUGUSED(event))

6 18

Filesystem.Tar()->get_dir() behaves oddly
by H. William Welliver III 20 Dec '12

20 Dec '12

I've been playing with a pluggable filesystem class that works with instances of Filesystem.Base, and have noticed that parts of Filesystem.Tar behave differently from Filesystem.System. In particular, paths expressed by Filesystem.Tar always seem to be full rather than the relative paths returned by Filesystem.System. That seems improper to me, especially if one expects Filesystem.Base()->get_dir() to behave in a manner similar to get_dir(). For example: > object f = Filesystem.Tar("foo.tar"); > f->get_dir(); (1) Result: ({ /* 2 elements */ "/Session.pmod", "/Tools.pmod" }) > f = f->cd("Session.pmod"); // note that the object is perfectly content to deal with relative paths > f->get_dir(); (3) Result: ({ /* 2 elements */ "/Session.pmod/RAMSessionStorage.pike", "/Session.pmod/SQLiteSessionStorage.pike", }) I've got a fix that seems to resolve this, but wanted to check that it's actually a bug (it seems as though it must be, as it would be hard for filesystem objects to be interchangeable if they behave differently in this way). Bill

1 0

Planned outage of pike.lysator.liu.se
by Henrik Grubbstr�m (Lysator) ＠ Pike (-) developers forum 18 Dec '12

18 Dec '12

It should be up now.

3 2

program names
by H. William Welliver III 11 Dec '12

11 Dec '12

I've got some code that generates pike program structs dynamically, and I was wondering if anyone could explain the mechanism that gives C-level programs their names when viewed from pike. For instance: object x = foo.MyProgram > object_program(x); (6) Result: dynamic_class.m:936 Where foo is a module which has `[] overloaded to generate c-api classes on demand. I assume that an entry would normally be created somewhere in the master, however because I'm bypassing all of that machinery, the necessary entries aren't being created. I assume that I should go back and create them, correct? If so, which tables really need to be populated? Bill

1 0

Assign where compare meant
by Peter Bortas ＠ Pike developers forum 02 Dec '12

02 Dec '12

Coverity picked up a few strange assignments in files/ that I commitet fixes for in 7.9. That they had been missed in the files module made me go hunting for more, and I found some in the signal handler. Does this make sense: diff --git a/src/signal_handler.c b/src/signal_handler.c index c886a86..b0a04bb 100644 --- a/src/signal_handler.c +++ b/src/signal_handler.c @@ -1076,7 +1076,7 @@ void forkd(int fd) do { j = close(i); } while ((j < 0) && (errno == EINTR)); - if ((j < 0) && (errno = EBADF)) num_fail++; + if ((j < 0) && (errno == EBADF)) num_fail++; } while (1) { @@ -1088,7 +1088,7 @@ void forkd(int fd) iov.iov_len = 1; do { i = recvmsg(fd, &msg, 0); - } while ((i < 0) && (errno = EINTR)); + } while ((i < 0) && (errno == EINTR)); if (!i) _exit(0); /* Connection closed, shutdown forkd. */ for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) { int ctrl_fd = NULL; @@ -1101,7 +1101,7 @@ void forkd(int fd) num_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int); do { i = fork(); - } while ((i < 0) && (errno = EINTR)); + } while ((i < 0) && (errno == EINTR)); if (i < 0) { /* Fork failure. */ } else if (i) { (Just in case those are there to acctually force the error to EINTR for some reason that I can't see.)

3 2

low_mega_apply switch
by Peter Bortas ＠ Pike developers forum 02 Dec '12

02 Dec '12

Is there a goot reason the default case for the big low_mega_apply switch is wedged in the middle and not at the end like normally done? If not I'm moving it.

1 0

Coverity Scan accounts
by Peter Bortas ＠ Pike developers forum 01 Dec '12

01 Dec '12

After dicussin it on the conferance I contacted Coverity and made a Pike project on their Open Source site. Those of you that want an account to check out and work on the defects it reports please reply to this with the user name and email address you want to use. If you don't want it in the list archives then send it to me directly at bortas(a)gmail.com.

1 1

Readline crash
by Martin Nilsson (Opera Mini - AFK!) ＠ Pike (-) developers forum 08 Nov '12

08 Nov '12

If I press ´ followed by pretty much any other button in hilfe I get Error decoding [0xb4] using utf8: Invalid byte. /usr/local/pike/7.9.5/lib/modules/_Charset.so:1: _Charset.UTF8dec()->feed("´") /usr/local/pike/7.9.5/lib/modules/Stdio.pmod/module.pmod:1815: Stdio.FILE("stdin", "r", 777 /* fd=0 */)->__lambda_66475_0_line_1814("´") /usr/local/pike/7.9.5/lib/modules/Stdio.pmod/module.pmod:1716: Stdio.FILE("stdin", "r", 777 /* fd=0 */)->low_get_data() /usr/local/pike/7.9.5/lib/modules/Stdio.pmod/module.pmod:2118: Stdio.FILE("stdin", "r", 777 /* fd=0 */)->read(1024,1) /usr/local/pike/7.9.5/lib/modules/Stdio.pmod/Readline.pike:582: Stdio.Readline()->InputController()->run_blocking() /usr/local/pike/7.9.5/lib/modules/Stdio.pmod/Readline.pike:1658: Stdio.Readline()->edit("",UNDEFINED,UNDEFINED) /usr/local/pike/7.9.5/lib/modules/Tools.pmod/Hilfe.pmod:2538: Tools.Hilfe.StdinHilfe()->create(UNDEFINED) /usr/local/pike/7.9.5/lib/modules/Tools.pmod/Hilfe.pmod:2474: Tools.Hilfe.StdinHilfe() Ignoring whatever is wrong with the terminal settings, where should this error be caught? It doesn't feel like something the user of the Readline object should have to care about, or is it?

1 0

sscanf bug?
by Martin Nilsson (Opera Mini - AFK!) ＠ Pike (-) developers forum 08 Nov '12

08 Nov '12

Shouldn't I get ({ "0", "1" }) from this? > array_sscanf("01", "%[0]s%s"); (1) Result: ({ /* 1 element */ "0" })

2 2

Jump to page:

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