Subject: publishing
chambers@cs.washington.edu
Date: Thu Oct 12 2000 - 13:44:55 PDT
I'm publishing my system, including a new *linux* vortex executable.
*) Keunwoo reported that the Linux version of Vortex was no longer
working (it couldn't read medium-sized input files, for one). I
worked on various hacky workarounds to the unix_file I/O code to try
to suppress the seg faults that Linux now seems to generate (and
handle) all the time, perhaps to do on-demand mapping of heap memory
pages, but which were screwing up fread calls. Things now appear to
work on Linux.
*) In the process, I discovered a bug where unix_file::read was
allowed to read part of its input and return the number of bytes read,
but that clients were assuming that all their requested input was
read. I fixed things by renaming read to read_partial, and adding a
read wrapper that calls read_partial in a loop to read all that's
requested.
*) At the same time, I changed the interface to unix_file functions so
that the error closure now takes a string, not an integer error code.
This allows functions to report arbitrary errors, not just those that
correspond to errno codes.
*) I did some patches to the implementation-side typechecking code to
work around some bugs. I think things still don't work fully, though,
since itc seems to hang in typechecking one GF. I'll look at it more
later.
*) I changed the list of linux machines to compile on in
compileMachineNames.
-- Craig
This archive was generated by hypermail 2b25 : Thu Oct 12 2000 - 13:45:03 PDT