Re: My schedule, 2.4 schedule

[Matthias Clasen <maclas gmx de>]

Am Di, den 23.12.2003 schrieb Owen Taylor um 17:17:
> My schedule is the following:
> 
>  Dec 23 => Dec 27
>    Holidays. I'll be checking mail and planning on doing 2.3.2
>    releases on Friday or Saturday.
> 
>  Dec 28 => Jan 4
>    Moving into new house / moved into new house. Presumably without
>    internet access. 
>  
>  Jan 5 
>    Back at work. Hopefully with GTK+ stuff in
>    January than I had this month.
> 
> http://gtk.org/plan/2.4/ needs updating again. What I'm thinking 
> is:
> 
>  Fri Jan 9  - Really final, hard, ultimate API/ABI freeze
>  Mon Feb 2 -  Final 2.4.0

I'll take care of this.

> Sound vaguely plausible?

Well, unless some patch review and approval takes place, all we'll have
on Jan 9 is a huge punt list... here is a random selection of GTK+
patches which I'm interested in and which need review:

API affecting
73207 gtk_accel_map_lock_path()
92346 Window icons don't respect icon theme
109213 GtkNotebook buttons in labels get highlighted on tab switch

not API affecting
107528 Menu keynav is backwards in RTL env
  (IMO a stopper for RTL support, which is otherwise shaping up nicely)
112412 Corrections done by correct_total seem excessive when using
INTERP_PIXOPS_HYPER
  (this one is nice to analyse offline while being without net
connection...)


And for convenience, here is my take on the <=2.4 GLib bugs again,
updated to the current state of affairs.
 
*: change ready to commit
+: bug I think should be fixed for 2.4.0
?: maybe fix, maybe punt  
-: good punt candidate
p: has a patch


GLib 2.2.4
==========

+p 119261 gspawn can hang when limits are set too high
   patch needs to be redone to incorporate Owens feedback

+  119490 system snprintf used even if broken
   should be easy, just tighten up some configure checks

?p 11059 Linux poll issue
   has a two years old patch which is said to need testing/debugging

?p 116617 still failed assertions for EINTR
   not clear if the patches are complete

?p 107626 g_module_symbol(NULL, ...) fails with loaded modules
   has a patch, but seems to require further thought and testing

?  82093 strsignal assumed to be either const or not - should be test
   Owen proposes a simple solution for 2.4 

?  98737 g_spawn* Win32 implementation needs work
   partially done, no idea how hard the remaining bits are

-  66199 Potential memory leak in gtk_init
   not touched for more than a year, seems to be blocked on disagreement
   between Owen & Tim

-  73861 iconv missing conversions on IRIX
   not touched for more than a year, open issue is to remove a warning
   if people start complaining about it

-  101248 MSVC makefiles need updates
   no idea what needs doing here

-  101792 g_dir_read_name: filename encoding on Win32
   not clear what to do here
   
-  118952 G_THREAD_LIBS used instead of G_THREAD_LIBS_FOR_GTHREAD in p
   toolchain/configuration problem on s390


GLib 2.4 API freeze
===================

*p 50296 UNIX signal main loop source
   has been through multiple iterations

*p 78414 g_queue_remove() and g_queue_remove_all()
   has been though multiple iterations

*p 88329 Add g_strtokenize()
   even has tests     

*p 98536 Add g_timer_continue()
   this one too

*p 121386 glib-mkenums: Allow override of lowercase-ization for _get_t
      
+p 129636 Seeding by array, and other seeding issues in GRand

+p 63621 Add GAtomicInt for atomic integer operations
   needs to be expanded to other architectures besides i486
   
+  107397 G_TYPE_BOILERPLATE wanted, as replacement of GNOME_CLASS_BOI
   should be easy
   (Owens comment from Sep 12: I think this is a *big* help in writing 
    new object types  and would really like to see this addition for 
    GObject-2.4.)

?p 121027 avoid emission of class closure only signals
   Tim wasn't convinced of the approach 

?p 51639 Equivalent of __FUNCTION__ for Sun's compiler
   patch needs to be redone to incorporate Owens feedback

?p 110528 collecting value arrays is inconvenient
   needs a better name than GStrv
   (Owens comment from Sep 12: Pretty trivial patch, main question is 
    whether NULL terminated arrays arrays like this are the right way 
    to do it.)

?  81040 G_TYPE_GTYPE is missing from fundamental types
   should be easy

?  99226 Add G_HIDDEN
   should be easy 
   (Owens comment from Sep 12: Might be worth looking at the 'slim' 
    stuff that Cairo is using for alternate symbol-hiding magic.)

-p 80498 Added function g_io_channel_new_pipe()
   needs a win32 implemenation to be useful

-p 111848 function to canonicalize file names
   needs to be modified to not follow symlinks, win32 implementation
   (Owens comment from Sep 12: I don't think we really have a good sense
    of exactly what we want here yet.)

-  83729 Patches to add iterators to GLib
   de facto WONTFIXed already, judging by the comments
   (Owens comment from Sep 12: I consider this a wontfix, but *still* 
    haven't written up the detailed rational for that)

-  103710 Why is there no G_MININT32/16/8, G_MAXINT32/16/8, G_MAXUINT3
   rather pointless constants

-  113075 support "nonnull" attribute
   not very useful

-  113697 g_ptr_array_sort_[with_data] is twisted
   the twistedness has at least been documented
   (Owens comment from Sep 12: I really don't know what to do with this 
    one. We need to come up with replacement names and deprecate the 
    current functions.)

-  122057 Add g_warning_location()
   
-  120299 Win32 iochannel/g_poll/etc code needs rewrite
   no patch, sounds like a lot of work


GLib 2.4
========

*p 118536 Make g_signal_connect_object'ed handlers disconnect when the
   patch has had some iterations

*p 118973 Pan does not understand "&" in username field of /etc/passwd
   implements a relatively obscure traditional passwd feature
   
+p 113116 Consider caching of type check information
   seems to provide a "not insignificant" speedup

+p 117067 Speed up g_param_spec_pool_lookup
         
?p 101426 GScanner is very slow with big files
   sitting there since December last year, waiting for a comment 
   from Tim...

?  106433 Unnecessary read lock in g_type_class_peek_parent
   should be trivial, according to Owen

-  106925 g_new0 performance
   should be simple

-  116308 Support ./configure XGETTEXT=blah
   sounds like a simple configure tweak   

-  71704 Header file include order
   purely janitorial

-  100697 Problems with -pthread and shared libraries
   just left open for removing some configure hacks when a fixed
   libtool appears one day

-  108038 Screwed-up headers guards
   purely janitorial

-  119615 glib-gettext.m4 obsolete macro
   not clear what to do

-  57691 check error conditions in backend read/write/seek
   no action on this for a long time; win32 specific