(see https://sympa.inria.fr/sympa/arc/ecm-discuss/2025-03/msg00000.html)

(https://sympa.inria.fr/sympa/arc/ecm-discuss/2025-03/msg00000.html)

(contributed by Dima Pasechnik <dima@pasechnik.info>)

bsave.B100.v99 is a save file with the fake version number 99 hex-edited in

If want_mmap!=0 and, somehow, HAVE_MMAP is undefined, return with an error.
This case is currently unreachable because caller sets want_mmap only when
HAVE_MMAP is defined, but if we test for this case here, then it should be
handled correctly.

This was suggested by Mehrshad Alipour

Including an s file created in the old format to test the "Wrong magic number"
message

Intended for the next minor version release.

Adds a new configure option: --enable-mmap
and, if mmap is enabled, two new command line options: -bsavems, -bloadms

The rationale is that when multiple ecm processes run on the same machine,
all doing ECM with batch stage 1 and the same B1, they all have their own
batch product s in memory which contains the same data for all processes.
As the batch product can become quite large with large B1, this may cause the
machine to run out of memory with many cores. By using a file-backed mmap
instead, the operating system can use shared memory between all processes.

The file format for the s save files has changed. It has a header now, with
magic number, endianness, B1 value and a checksum. The file format is
different between mmap-ed and non-mmap-ed data; the mmap-ed one uses the
host's endianness for mp_limb_t (as the disk file is just a memory image)
while the non-mmap-ed one uses mpz_out_raw() (as before) which always
writes in big-endian.

with --enable-assert, assertions were not enabled

It might fail, for example with "Error, -bsaves makes sense in batch mode only"
error, and then all following tests also exhibit that error - which is not
detected, because those tests are expected to return an error code.

If the random sigma value fails, then try the next one (sigma+1),
and so on, instead of drawing a new random value. For small n,
this will avoid failures (for n=5, the probability of failure
with param=1 goes down from 1/150 to 0).