After a few private discussions with KernelTraffic, I decided to keep this gcc newsletter independent: R. D. Flowers offered a host and bandwidth on the new host http://gccnews.chatta.us. As such, it should stay there for a while.
Mark Mitchell sent a small report on the status of gcc 3.4.1 which was supposedly released on June the 15th.
The merge of the Objective C++ frontend contributed by Apple has been officially approved although no deadlines were discussed.
The status of the merge of the LNO branch was summarized by this email from Zdenek Dvorak.
The removal of the expect and the dejagnu directories proved to be hot topic: it was discussed until last thursday. Ben Elliston who originally started the thread seemed to give up but later came back with a new strength.
Since the job of an optimizing compiler is to optimize, it comes as no surprise that the compiler needs a measure of the performance of the generated code to compare multiple versions of the same code and keep only the best one. I found this discussion on what metric to use and how to implement it interesting. The last mail sent by Toshi describes an interesting possible solution.
That kind of mail is the typical flamewar food but none really started. I would have thought the followup would trigger something serious but no one started jumping around with crazy eyes. I am a bit disapointed.
Since the last gcc newsletter, I discovered a few new summaries discussing what happened during the gcc summit. Most notably, Dan Kegel's notes.
The gcc summit also prompted another discussion after Ben Elliston posted Notes from the testing BOF. The discussion focused on parallel testing to decrease testing time with dejagnu, the testing framework used by gcc. This prompted the proposal to remove from the cvs repository the expect and dejagnu directories which contained stale copies of these projects. The recommended way to use these is now to install them yourself on your host system although they have not been removed yet from the CVS repository.
Daniel Berlin seems to be on the way to improve the regression reporting by auto-testers: his new tool looks awesome.
Richard Kenner confirmed that the conversion of the Ada frontend to the tree-ssa architecture is underway. No schedule was detailed but the continuous stream of questions he posts on how to use tree-ssa implies he is hard at work.
Badley Lucier is pushing to merge the new-ra branch into mainline. Since no one clearly spoke against it, this might happen soon but it probably will have little impact on most users since it merely improves the -fnewra compile switch which enables the so-called New Register Allocation. I have been unable to find a homepage for this project so, I cannot comment much on it.
New optimizations made possible by the new tree-ssa framework are coming in. A preliminary version of IPA constant propagation was sent to the list.
Some gcc users sometime try to do really weird things. One of these is C++ exceptions in UNIX signal handlers.
Mark Mitchell keeps working on gcc 3.4.1. The release status as of the 31-05-2004 is summarized in this mail. The release date seems to be still June 15th.
Work on cvs HEAD (aka gcc 3.5) is still ongoing: stage 1 is about to be closed on July 1st.
I am sure a lot of fruitful discussions have taken place at the GCC summit. For now, the complete set of papers presented there is available there.
There is nothing like a good old discussion on the version number of the next major version of gcc to start a day. Some argue for gcc 4.0. Others for gcc 3.5. Because I don't want to get into a flamewar, I won't make any comments about this one :). It was started all by this email as a followup to previous discussions on the same topic a few weeks ago.
Because I did not know about most of these projects (I toyed with similar ideas but did not know a lot of people had already done it at least partly), I thought I should mention this low-volume thread on Omniscient Debugging started by this mail. A number of replies pointed to previous work in this area, including the gdb's feature named tracepoints but also Lizard.
It was bound to happen someday: a discussion on version control tools was started by this email. It evolved quickly into a non-flamewar :)
A lot of questions on the layout of Gimple trees and how to use them were asked by Richard Kenner. Maybe he has started working on porting the Ada95 frontend to the new tree-ssa architecture...
Richard Guenther seemd psyched by a 10% compile-time decrease as shown on his automatic tester.