[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
https://issues.dlang.org/show_bug.cgi?id=3637 Andrei Alexandrescu changed: What|Removed |Added Version|future |D2 --
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 Walter Bright changed: What|Removed |Added Status|ASSIGNED|RESOLVED CC||bugzi...@digitalmars.com Resolution||FIXED --- Comment #9 from Walter Bright 2010-03-08 22:26:33 PST --- Fixed dmd 2.041 -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 Steven Schveighoffer changed: What|Removed |Added Status|NEW |ASSIGNED AssignedTo|s...@invisibleduck.org |schvei...@yahoo.com --- Comment #8 from Steven Schveighoffer 2010-02-22 08:37:48 PST --- changeset 252 http://www.dsource.org/projects/druntime/changeset/252 -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 Steven Schveighoffer changed: What|Removed |Added Attachment #569 is|0 |1 obsolete|| --- Comment #7 from Steven Schveighoffer 2010-02-18 06:34:43 PST --- Created an attachment (id=570) Patch to druntime revision 245 to implement array appending The only addition to this patch is I added comments identifying where the compiler fails to deliver the proper typeinfo for shared appending. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 --- Comment #6 from Steven Schveighoffer 2010-02-18 06:29:32 PST --- Note that shared data isn't properly synchronized when being copied, but this also happens in all the other array functions (not just appending). I will file a separate bug on this. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 Steven Schveighoffer changed: What|Removed |Added Attachment #528 is|0 |1 obsolete|| --- Comment #5 from Steven Schveighoffer 2010-02-18 06:28:15 PST --- Created an attachment (id=569) Patch to druntime revision 245 to implement array appending generated with svn diff. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 --- Comment #4 from Leandro Lucarella 2009-12-23 09:33:20 PST --- (In reply to comment #3) > It's not *necessary* to change the API to fix this, but it is hugely > advantageous. If you want to know the block size of the chunk you just > allocated, the current API requires *another* lock of the GC, and a search > through the pools. All the info is there in malloc, it's just not returned. > > BTW, just an additional size parameter would suffice (this is how gcx handles > the allocation). Having it return a BlkInfo struct is convenient because that > is the data type I'm working with when setting allocated length. Yes, that seems reasonable since BlkInfo is already part of the API. > And introducing a new function is just as backwards compatible as adding a new > optional parameter to a current function. I know that, but it adds complexity to the API, and seems a little weird since gc_malloc() and gc_malloc_bi() are basically the same. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 --- Comment #3 from Steven Schveighoffer 2009-12-22 20:19:09 PST --- It's not *necessary* to change the API to fix this, but it is hugely advantageous. If you want to know the block size of the chunk you just allocated, the current API requires *another* lock of the GC, and a search through the pools. All the info is there in malloc, it's just not returned. BTW, just an additional size parameter would suffice (this is how gcx handles the allocation). Having it return a BlkInfo struct is convenient because that is the data type I'm working with when setting allocated length. And introducing a new function is just as backwards compatible as adding a new optional parameter to a current function. Preallocation needs to be a runtime function that is yet to be written which allocates a large block but sets the "allocated" size to 0. It can also avoid pre-initializing the block if the block has no pointers (something that the old trick didn't do). -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 Leandro Lucarella changed: What|Removed |Added CC||llu...@gmail.com --- Comment #2 from Leandro Lucarella 2009-12-22 15:55:14 PST --- It is really necessary change the GC API to fix this? I think it's a really bad idea to modify the GC API (add a function) just because dynamic arrays have problems (I don't want to start over the discussion about dynamic arrays being broken :). If you absolutely need this new function: BlkInfo gc_malloc_bi(size_t sz, uint ba = 0); then maybe it's better to change the existing regular gc_malloc() to: void* gc_malloc(size_t sz, uint ba = 0, BlkInfo bi = null); The advantage is that it's API-backward compatible and doesn't introduce a new function to the GC API and the downside is it's not binary-backward compatible, but I don't think people care much about this in D. The inability to pre-allocate can be a little bad too, but it was awkward anyway, so providing a better way to do so can be a good thing after all. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---
[Issue 3637] Array append patch to prevent stomping and to enhance thread-local append performance
http://d.puremagic.com/issues/show_bug.cgi?id=3637 Leandro Lucarella changed: What|Removed |Added Attachment #528|application/octet-stream|text/plain mime type|| Attachment #528 is|0 |1 patch|| --- Comment #1 from Leandro Lucarella 2009-12-22 15:35:51 PST --- (From update of attachment 528) Fix the MIME type and mark the attachment as a patch. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email --- You are receiving this mail because: ---