We had a meeting of the Mercury group on Fri Mar 2, 2001, from 3.15 to 4.00pm. Attendees: Fergus Henderson (fjh) Zoltan Somogyi (zs) David Jeffery (dgj) Simon Taylor (stayl) David Overton (dmo) Thomas Conway (conway) Kevin Glynn (keving) Adrian Pellas-Rice (apell) 1. Background We have recently acquired two more machines, miles, a laptop, and earth, a workstation. Simon Taylor (stayl) has returned to the fold. The summer students being supervised by Zoltan (zs) have left the office, although some are continuing their work remotely. 2. Progress reports We went around the table with each person in turn discussing what they had been working on recently and/or what they were planning to work on. Fergus Henderson (fjh) - some further work on the gcc back end. At present, installing libraries using mmake with the new back end does not work. - merging in Robert (rejj) Jeschofnik's changes to allow the expansion of equivalence types to work with the .NET and Java backends. Fergus expressed some uncertainty about how best to go about these changes. He then debated at length with Zoltan (zs) Somogyi about whether/where equivalence types should occur in .int* files. - plans a similar change for the handling of abstract equivalence types. Zoltan suggests that abstract equivalence types should be expanded no earlier than code generation, however Fergus was concerned that this would be too hard. Zoltan offered an additional about idea using abstract maps for this task. Zoltan and Fergus agreed that this proposed change is likely to conflict with Zoltan's current work. David Overton (dmo) - has continued work on mode checking. Is experimenting with different representations, particularly hash tables and bitsets. Tom (conway) Conway mentioned that in David Overton's case, specialisation of the calls to the library's hash table module are highly desirable. David (dgj) Jeffery pointed out that the storage of the hash function in the hash table data structure makes this a particularly tricky task. Zoltan Somogyi (zs) - has been fixing deep profiling bugs. With these changes, deep profiling now bootchecks. - posted the diff for liveness this week. It bootchecks in two grades and has been committed. - Has improved the accurate garbage collector so that it now succeeds to the second stage of a bootcheck. (It still bombs out in stage three, of course.) David Jeffery (dgj) - has been working on his thesis. A draft will be ready presently. Kevin Glynn (keving) - At this point Kevin revealed that he was attending the Mercury meeting for stress relief. It may not have killed him yet, but after that remark, Kevin's thesis is clearly approaching the potency of ebola. Simon Taylor (stayl) - has attended to software installation on earth. (Recall that `earth' is our new machine.) - part way to fixing a performance bug in deforestation. Thomas Conway (conway) - has been working on his thesis. - collaborated with Zoltan (zs) Somogyi on deep profiling. - plans to return to MDS within 3 weeks. Adrian Pellas-Rice (apell) - continued the well-behind-schedule implementation of sequence quantification. - had some further assistance from Fergus (fjh) Henderson pertaining to a serious bug. 3. Administrative Matters: - Zoltan (zs) Somogyi complained about the frequency of the emails requesting bug fixes. The present frequency is one email per bug per day. After some debate, Fergus (fjh) Henderson agreed to change the frequency to the compromise position of one email per bug per three days. - On this note, Fergus also suggested splitting the [mdev] mailing list into two lists: one for diffs and one for all other correspondence. Zoltan advocated an even more radical split into several topic based lists. Discussion will resume at a later date. - Zoltan requested that more file systems be exported to miles (the laptop), and requested assistance with some of the more concisely documented software on miles. Simon (stayl) Taylor agreed to see to the file systems, David (dmo) Overton agreed to see to the software. 4. General Matters: - Fergus and Zoltan discussed a problem with value numbering. Though a temporary fix is in place, Zoltan advocates the removal of value numbering as the long term repair strategy. - There were some exchanges about the grades and modules that are approaching redundancy. Fergus wanted to dispose of the transient register windows soon, but Zoltan would prefer to wait until the high level back end supports debugging. There was general agreement that supporting high performance on the SPARC is no longer a priority, and possible strategies for reducing the burden of the SPARC legacy were mentioned. - Zoltan enquired about need to retain the old profiler once the new (deep) profiler is completed. Tom (conway) Conway declared that the old profiler should be kept because a) in some cases it incurs less overhead than the new profiler and b) it is easier to do time profiling with the old profiler. Zoltan observed that retaining both profilers would be a great hassle, but said he would consider some of the required changes. 5. The 0.10 Release: - The 0.10 release has proceeded, with an announcement on mercury-announce. - Fergus expressed his desire to freeze 0.10.1 next week, and reminded everyone to commit bug fixes on the main branch too. - The release of 0.10.1 will be announced through the major channels as well as mercury-announce. Minutes taken by Adrian Pellas-Rice (apell@cs.mu.oz.au).