temp buffers about nvc HOT 4 CLOSED

Are they any particular fixed-sized buffers which are causing problems? Some of the ones in src/link.c and src/rt/jit.c could be replaced with xasprintf and a corresponding free (wrapper in src/util.c for asprintf that aborts on failure).

I looked at libgc before but it's way over the top for this use. It causes all sorts of weird non-deterministic bugs and more importantly, is incompatible with Valgrind.

AddressSantitizer looks quite good as a memory checker and if using Clang it only needs an extra compiler flag to enable.

from nvc.

for fixed-sized buffers, maybe i can spoil them by using crazy VHDL source file which has very long identifier of entity (or file name?). IIRC, LRM does not mention about limitation about length of identifier. it looks you already set reasonable length for them, and i also expect sane users do not use very long identifier, but some HLS tools generate crazy ones, such as Vivado.
http://www.xilinx.com/support/answers/52787.htm
in fact, the above one is caused by Windows specific MAX_PATH limitation (I believe this can be fixed by using namespaces though). but this craze of HLS will produce more brilliant shits in near future, IMHO.

using xasprintf looks nice for char, ill try it soonish.

for GC and memory checker, it looks you already have your own thoughts, so i have no objection about them.

P.S. ive updated the top post.

from nvc.

commit a0a9ba0 purges 20 temp buffers.

any idea for nuking others?

from nvc.

Thanks, I merged that commit. All fine except two little issues:

In _bounds_fail the suffix variable always gets passed to printf so it has to be non-null or you end up with error messages like this:

** Fatal: array index 9 outside bounds 8 downto 0(null)

Fixed that in ab89cd8. It's OK to not free suffix as _bounds_fail never returns.

In loc_to_ident it's possible we write past the end of the array. This function has always been kind of broken though. Fixed it with 462a33d.

I also changed a bunch of fixed size buffers in cgen.c to use xasprinf too in 90cfa27. I don't think the remaining ones are anything to worry about at the moment.

from nvc.