upx/TODO
Markus F.X.J. Oberhumer a77b598f82 Implement a micro-optimization which has been on our TODO list for ages:
remove "or ebp,-1" when not needed
2006-12-08 09:20:14 +01:00

115 lines
2.3 KiB
Plaintext

UPX TODO list. Last updated 2006-12-08.
IMPORTANT PROBLEMS THAT SHOULD BE FIXED SOON:
- [None]
OTHER:
- docs: convert docs from upx.pod to use AsciiDoc
- check all <const_cast> to make sure they are not invalid
- throwNotCompressible() is not a real error, so make the output nicer
(info: bla bla). Also ui.cpp (total_*).
-----------------------------------------------------------------------
IMPROVED COMPRESSION RATIO
==========================
- experiment with new filters
- implement filters for dos/exe
- filters: could we exploit a f->firstcall info field ?
- for small programs (e.g. < 64k), try an additional algorithm
to see if it gives better compression
ALL FORMATS
===========
- more thoroughly test the exe-header in canPack()
and throw exceptions when encountering bad values.
- implement `--cpu=486' option to use bswap on the 32-bit formats
(if cpu >= 486)
FORMAT DJGPP2/COFF
==================
- handle overlays
- fix default file extension handling when the --coff option is set
FORMAT DOS/EXE
==============
- implement filters
- add a check so that we don't pack djgpp1 binaries
FORMAT LINUX/386
================
- don't mmap() the temporary output file - this seems to improve
file io speed
FORMAT TMT/ADAM
===============
- the decompressors are already aligned, no need for an
extra alignment
FORMAT WATCOM/LE
================
- handle files without relocations
- the decompressors are already aligned, no need for an
extra alignment
- fix default file extension handling when the --le option is set
- handle holes in the file
FORMAT WIN16/NE
===============
- implement readFileHeader() to correctly identify a win16/ne
executable, so that the call for contribution will get thrown
FORMAT WIN32/PE
===============
- fix the section alignment with the Intel compiler
- decrease runtime memory overhead
3 - difficult) don't compress the BSS section and other holes.
4 - medium - ml) fix when objectalign < 0x1000
4 - easy - ml) put the original offset of moved resources somewhere into
the res.dir. (if it's safe to do)
4 - ??? - ml) fix FIXMEs
5 - medium - ml) try to put the preprocessed imports & relocs back to their
original section if possible. this could save some virtual memory
address space.