54 lines
1.7 KiB
Plaintext
54 lines
1.7 KiB
Plaintext
Changes to CGEN:
|
|
|
|
4/9/92 ---------- Released v5.2a beta 1 ---------------------
|
|
|
|
9/2/92 trailer() *jd
|
|
Added +t option to force $$var's (used for far data items) into codeseg
|
|
instead of dataseg.
|
|
|
|
9/11/92 starregs() *jd
|
|
Bad code generated for:
|
|
int a; huge struct mystr {int i,k,j;} *mp;
|
|
m() { a = mp->j; a = mp->i; }
|
|
|
|
9/14/92 gencast() *jd
|
|
When a null ptr was cast from near to far, result was not null (it was ds:0)
|
|
|
|
9/15/92 jd: outoper()
|
|
For the following:
|
|
(*(far void(*)())0x12345678L)()
|
|
cgen generated far call to 5678h, not 0x12345678.
|
|
|
|
9/16/92 jd: genstr()
|
|
For some structure copying, cgen generated bad code, because it
|
|
always prepared right side first, even if left was more complex.
|
|
Changed cgen to do hardest side first.
|
|
|
|
9/16/92 jd: templ.inp
|
|
For addition of constant to a huge pointer, compiler wasn't generating
|
|
extrn references to $hadd. For inc of huge ptr, it wasn't generating
|
|
extrn ref to $hinx.
|
|
|
|
|
|
10/4/92 ---------------- Released v5.2a beta 2 ----------------------------
|
|
|
|
|
|
10/5/92 jd: execvp()
|
|
Relinked 'cgen' to use fixed execvp() - previous execvp() didn't always
|
|
correctly search all directories specified in PATH.
|
|
|
|
10/22/92 jd: EvalRightFirst()
|
|
For some expressions, cgen was dereferencing a null pointer, causing
|
|
random areas of memory to be modified
|
|
|
|
10/22/92 jd: gencast()
|
|
The check for a null pointer in a cast of a near pointer to far
|
|
does not always have to be done. cgen was modified so that if the
|
|
near pointer is guarenteed not to be null, the check is not done,
|
|
thus saving space.
|
|
|
|
10/23/92 jd: outstatic()
|
|
cgen was putting near static uninitialized variables in the data
|
|
segment, not the bss segment, thus increasing the size of the data
|
|
seg.
|