home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ST-Computer Leser 2002 January
/
STC_CD_01_2002.iso
/
SYSTEM
/
XAAES936
/
DOCS
/
TODO.TXT
< prev
next >
Wrap
Text File
|
2000-09-06
|
1KB
|
44 lines
Progdef drawing under GCC must not destroy any registers
except d0/d1/a0/a1. Make sure! HR: see progdef.s
Check MiNT bugs (pipe open etc).
Can evnt_multi() be direct call?
Don't use malloc()!
One pipe per application?
240 kbyte client array? HR: fixed
Why one pipe/app? Limit to 32 apps!?!
Why would anyone want sockets instead of pipes?
HR: Pipes are effective and
easy to use and understand.
Handle indirect ob_spec? HR: seems to be OK
Fix progdefs 100% under GCC (register a2) (register d2)
HR: see progdef.s ?
Something must _really_ be done about the mouse handling.
HR :-( So many spurious (), and the only pair that matters is missing !!!!
if (flag & 1 == 0) --> ((flag & 1) == 0)
Only after I tidied up the complex if's using vertically aligned
parenthesis', I saw it.
in C binary & is weaker than ==
this result in implicit bracing: if (flag & (1 == 0))
1 == 0 yields 0, so setting the flag is never detected. Voila ;-)
Now I must check every & in the whole source tree.
There's no synchronization between evnt_multi() and the message send code.
If an evnt_multi() calls occurs at the wrong moment, it might miss a
messages. Something similar could happen with at least mouse events too.
Use the KBD device
There seem to be some kind of semaphore race condition since sometimes
the root window icons are redrawn on top of other windows.