The mechanisms to use pre-generated source for BSDL and SVF from distribution tar.gz don't seem to work perfectly
I just ran a "make dist" and tried to ./configure the result on a Cygwin system.
SVF: It just detected the 2.5.4 flex and disabled the SVF system.
BSDL: The BSDL subsystem remained enabled, but can't be compiled; maybe the bsdl_sysdep.h has to be added to the distribution?
bsdl_bison.y:131:25: bsdl_sysdep.h: No such file or directory
Logged In: YES
user_id=478715
Originator: YES
"disabled the SVF system. " - sorry, I meant "disabled support for printing error locations".
Logged In: YES
user_id=156321
Originator: NO
The bsdl_sysdep.h issue should be resolved with r1040.
Need more investigation for the SVF flex detection.
Logged In: YES
user_id=156321
Originator: NO
When I implemented the check for proper flex version with BSDL (ref. [ 1873515 ]), I though about having the same logic for the bison locations. But that would break things under certain conditions. Consider the following scenario if bison locations check would be implemented like for BSDL:
* user checks out from SVN
* ./configure detects old flex
-> bison locations disabled
* user compiles code
-> results in SVF player (static and dynamically created files) without locations
* user does ./configure once again
-> configure detects "released package"
-> bison locations enabled
* user recompiles code (with unmodified svf_flex.l)
-> results in inconsistent SVF player
static files have locations enabled (via #ifdef)
dynamically created files have it still disabled (flex output from first compile)
This was where I stopped back then.
I agree that the current situation is suboptimal but I don't have a solution that covers all cases.
There's some magic required so that re-running ./configure doesn't think the user builds from a released package and still keeps bison locations disabled.
Logged In: YES
user_id=478715
Originator: YES
The BSDL part works now, thanks. Regarding SVF:
I thought that all information about flex version is contained in the auto-generated svf_flex.c itself, rather than being provided by configure (e.g. in config.h).
If that's true, it would be sufficient (and I thought it was working that way) to just make sure that svf_flex.c isn't regenerated unless there's a good reason for it?
Logged In: YES
user_id=156321
Originator: NO
Please check whether r1042 works as expected:
- keep bison locations status "as is" when svf_flex.c is existing
- disables bison locations upon virgin SVN checkout
Logged In: YES
user_id=478715
Originator: YES
Looks good, thank you!