Subj : Correction to previous - ticket #9 stuff
To : All
From : Sampsa Laine
Date : Sat Oct 29 2016 10:54 pm
Hi Guys,
First of all, let me say that I'm a HUGE fan of MBSE, have been for literally
for almost as long as I've been playing with UNIX BBSing (early 90s onwards).
Anyway, I posted a bug report a while ago and was wondering if anyone
actually
took a look at it?
TL;DWR: The bug is at
https://sourceforge.net/p/mbsebbs/tickets/9/ (ticket
#9,
damn it, I WANTED A PRIME NUMBER :P) and my more indepth explanatin of the
problems I encountered are here:
Not sure if you guys still read this echo but I'm running MBSE 1.0.6 and I
can across a few bug and ssome minor inconveniences in the compilation
process,
mainly:
I. There are some places in the code where a pointer is case to an int.
This DOES NOT work on certain platform, zLinux being one of them with
32-bit ints and 64-bit pointer.
For example, this is potentially dangerous:
filelist.c:496:3: warning: cast from pointer to integer of different size
\
[-Wpointer-to-int-cast (unsigned int)tmpf->flofp, (unsigned
int)tmpf->floff);
Luckily there are only TWO of these bugs in the whole codebase, maybe
some
leftover from the everything-is-32-bit era?
II. If certain #ifdefs do not recognise the CPU, naturally the code doesn't
know to call it and what the signal names on that platform are. IMHO this
could be solved by either:
- Having an #elseif, throwing a #warning about unsupported CPU and
setting
the signal name array to some sensible default (I used x86, have no
issues)
OR
- Passing in a -DCPU_OVERRIDE_ID="FooProc" and
-DUSE_SIGNALS="SomeOtherCpu",
and instructing the user todo this.
My suggestions as to fixes?
I. Never cast a pointer to an int or vice versa. This occurrs exactly TWICE
in your code base, filelist.c:496 and ftn2rfc.c:308 so not a huge fix I
think.
II. Dropping necessary(IMHO) #error's if the CPU is "not one of the above"
In a number of places, the code does a #ifdef to determne the CPU type and
whilst I'm impressed by the number of platform supported IBM S/390 and/or
z/Series aren't represented (not that I'm surprised, I'm probably the only
person in the world running MBSE on any kind of zLinux.
Anyway instead of #error'ing out of the whole compile, I changed the bits
which are CPU sensitive to instead throw a #warn "Unknown CPU, assuming IBM
S/390" [of course this should probablysay some differnet].
So from lib/strutil.c:399:2: error: #error "Unknown CPU" I do:
File: lib/strutil.c, starting at line 398>
398: #else
399: #warning "Unknown CPU - defaulting to IBM S/390"
400: return (char *)"IBM S/390";
401: #endif
402: }
I did these/similar changes to these files:
lib/signame.c
lib/strutil.c
mbmon/mbmon.c
mbtask/signame.c
(I just used the signal name array for x86, not ideal but I really didn't
have the time to look up the exact Debian/s390x versions - this shouldn't be
too strenuous)
Again you could do this with a -DCPU_OVERRIDE_ID="FooProc" type flag if you
don't think an #elseif is a good idea.
Conclusion
----------
On the plus side, I've been running MBSE 1.0.6 on Debian/s390x for months at
a time without a problem, so nothing too vital seems to be broken.
I would however suggest that you look into those void*<->int casts because
that could break on other platforms as well..
--- SBBSecho 2.27-Win32
* Origin: B4BBS = London = b4bbs.sampsa.com 2:250/7 (2:250/7)
