Re: PDUMP vs. UNION_TYPE + C++
On Fri, Jan 21, 2000 at 02:36:03AM -0800, Martin Buchholz wrote:
1. Since this is called the _portable_ dumper, I suggest you use
standard I/O from stdio, rather than low-level Unix I/O. In
practice this means we should use fwrite instead of write, etc.
This will also be much faster for creating the dump file, not that
that is very important (buffering).
Agreed.
2. XEmacs crashes for me as follows:
[...]
What I don't understand is how the lrecord_implementations_table
can
possibly be resurrected in the post-dump temacs. It seems like all
the pointers in the lrecord_implementations_table need to be
relocated, and perhaps the structures they point to need to be
re-created. I don't see where this is being done. I must be missing
something.
The structures they point to are global variables and global variables
aren't supposed to move around...
Could you compare the position in the file when dumping the
lrecord_implementations_table (done in pdump_dump_itable(), you can
get the position with a lseek(pdump_fd, 0, SEEK_CUR)) with the one
when loading (in pdump_load, just before the memcpy(), by computing
p-pdump_start) ?
If they are identical, I'm baffled. If they differ, the problem lies
in the previous reconstructions.
Olivier, you can get access to my machine for testing, if you need
it,
by sending me your ssh public key.
1024 37
94581611712733882108603919973713340704059786158843216569173003979560159045828004287494301131794196592462220896195338343069600041476710618653042836847217596019673319690898194810258685755173477694636267116128432338416027645439260800883971066536815085116041686293289687123131842650745179573382288419348564917299
galibert＠iria
OG.