todo

开源的nasm编译器源码,研究编译器原理很有帮且

NASM TODO list
==============

This, like the AUTHORS file, is intended for easy readability by both human 
and machine, thus the format.

  F: feature
  V: version you should expect it by
  R: responsible person or - if unassigned
  C: % complete
  D: description
  D: maybe on multiple lines
  
Anything that doesn't start with /^[FVRCD]:/ should be ignored.

  F:-line triggers new entry.
  Empty V,R,C assume: V: ?, R: -, C: 0%

=============

F: i18n via gettext

F: Convert shallow code model to deep code model
D: Tired of messing between lots of unrelated files (especially .c/.h stuff)

F: Automated dependency generation for Makefile
D: Current looks awful and will break if anything changes.

F: Move output modules out*.c to output/ subdir
R: madfire
C: 10%

== THESE ARE FROM old NASM's Wishlist
== THEY NEED SEVERE REVISING (seems they weren't updated for a couple of years or so)

F: Check misc/ide.cfg into RCS as Watcom IDE enhancement thingy
V: 0.98
D: ([e-mail])

F: Package the Linux Assembler HOWTO
V: 0.98

F: 3DNow!, SSE and other extensions need documenting
V: 0.98
D: hpa: Does it really make sense to have a whole instruction set
D: reference packaged with the assembler?

F: prototypes of lrotate don't match in test/*. Fix.
V: 0.98

F: Build djgpp binaries for 0.98 onwards. Look into PMODE/W as a stub
V: 0.98
D: it might be a lot better than CWSDPMI. It's in PMW133.ZIP.

F: %undef operator that goes along with %define
V: ?
C: 100%

F: Fix `%error' giving error messages twice.
V: 0.99
D: Not especially important, as changes planned for 1.1x below will make
D: the preprocessor be only called once.

F: Sort out problems with OBJ
V: 0.99
D: * TLINK32 doesn't seem to like SEGDEF32 et al. So for that, we
D:   should avoid xxx32 records wherever we can.
D: * However, didn't we change _to_ using xxx32 at some stage? Try
D:   to remember why and when.
D: * Apparently Delphi's linker has trouble with two or more
D:   globals being defined inside a PUBDEF32. Don't even know if it
D:   _can_ cope with a PUBDEF16.
D: * Might need extra flags. *sigh*

F: Symbol table output may possibly be useful.
V: 0.99
D: Ken Martwick ([e-mail]) wants the following format:
D: 	labelname	type	offset(hex)	repetition count
D: Possibly include xref addresses after repetition count?

F: ELF fixes
V: 0.99
D: There are various other bugs in outelf.c that make certain kinds
D: of relocation not work. See zbrown.asm. Looks like we may have to do
D: a major rewrite of parts of it. Compare some NASM code output with
D: equivalent GAS code output. Look at the ELF spec. Generally fix things.

F: ELF fixes
V: 0.99
D: NASM is currently using a kludge in ELF that involves defining
D: a symbol at a zero absolute offset. This isn't needed, as the
D: documented solution to the problem that this solves is to use
D: SHN_UNDEF.

F: Debug information, in all formats it can be usefully done in.
V: 0.99
D: * including line-number record support.
D: * "George C. Lindauer" <[e-mail]>
D:   wants to have some say in how this goes through.
D: * Andrew Crabtree <[e-mail]> wants to help out.

F: Think about a line-continuation character.
V: 0.99

F: Consider allowing declaration of two labels on the same line,
V: 0.99
D: syntax 'label1[:] label2[:] ... instruction'.
D: Need to investigate feasibility.

F: Quoting of quotes by doubling them, in string and char constants.
V: 0.99

F: Two-operand syntax for SEGMENT/SECTION macro to avoid warnings
D: of ignored section parameters on reissue of __SECT__.
D: Or maybe skip the warning if the given parameters are identical to
D: what was actually stored. Investigate.				
V: 0.99

F: Apparently we are not missing a PSRAQ instruction, because it
D: doesn't exist.  Check that it doesn't exist as an undocumented
D: instruction, or something stupid like that.
V: 0.99

F: Any assembled form starting 0x80 can also start 0x82.
V: 1.00
D: ndisasm should know this. New special code in instruction encodings, probably.

F: Pointing an EQU at an external symbol now generates an error.
V: 1.05
D: There may be a better way of handling this; we should look into it.
D: Ideally, the label mechanism should be changed to cope with one
D: label being declared relative to another - that may work, but could be
D: a pain to implement (or is it? it may be easy enough that you just
D: need to declare a new offset in the same segment...) This should be done
D: before v1.0 is released. There is a comment regarding this in labels.c,
D: towards the end of the file, which discusses ways of fixing this.

F: nested %rep used to cause a panic.
V: 1.10
D: Now a more informative error message is produced. This problem whould
D: be fixed before v1.0.
D: See comment in switch() statement block for PP_REP in do_directive()
D: in preproc.c (line 1585, or thereabouts)

F: Contribution
D: zgraeme.tar contains improved hash table routines
D: contributed by Graeme Defty <[e-mail]> for use in the
D: label manager.

F: Contribution
D: zsyntax.zip contains a syntax-highlighting mode for
D: NASM, for use with the Aurora text editor (??).

F: Contribution
D: zvim.zip contains a syntax-highlighting mode for NASM, for use with vim.

F: Contribution
D: zkendal1.zip and zkendal2.zip contain Kendall
D: Bennett's (<[e-mail]>) alternative syntax stuff,
D: providing an alternative syntax mode for NASM which allows a macro
D: set to be written that allows the same source files to be
D: assembled with NASM and TASM.
R: Kendall Bennett
C: 100%

F: Add the UD2 instruction.
C: 100%

F: Add the four instructions documented in 24368901.pdf (Intel's own document).
C: 100%

F: Some means of avoiding MOV memoffs,EAX which apparently the
D: Pentium pairing detector thinks modifies EAX. Similar means of
D: choosing instruction encodings where necessary.
V: 1.10?

F: The example of ..@ makes it clear that a ..@ label isn't just
D: local, but doesn't make it clear that it isn't just global either.

F: hpa wants an evaluator operator for ceil(log2(x)).

F: Extra reloc types in ELF
D: R_386_16 type 20, PC16 is 21, 8 is 22, PC8 is 23.
D: Add support for the 16s at least.

F: Lazy section creation or selective section output
D: in COFF/win32 at least and probably other formats: don't bother to emit a section
D: if it contains no data. Particularly the default auto-created
D: section. We believe zero-length sections crash at least WLINK (in win32).

F: Make the flags field in `struct itemplate' in insns.h a long instead of an int.
C: 100%?

F: Implement %ifref to check whether a single-line macro has ever been expanded since (last re) definition. Or maybe not. We'll see.

F: add pointer to \k{insLEAVE} and \k{insENTER} in chapters about	mixed-language programming.

F: Some equivalent to TASM's GLOBAL directive
D: ie something which defines a symbol as external if it doesn't end up being defined
D: but defines it as public if it does end up being defined.

F: Documentation doesn't explain about C++ name mangling.

F: see if BITS can be made to do anything sensible in obj (eg set the default new-segment property to Use32).

F: OBJ: coalesce consecutive offset and segment fixups for the same location into full-32bit-pointer fixups.
D: This is apparently necessary because some twazzock in the PowerBASIC development
D: team didn't design to support the OMF spec the way the rest of the
D: world sees it.

F: Allow % to be separated from the rest of a preproc directive, for	alternative directive indentation styles.

F: __DATE__, __TIME__, and text variants of __NASM_MAJOR__ and __NASM_MINOR__.

F: Warn on TIMES combined with multi-line macros.
V: 1.00
D: TIMES gets applied to first line only - should bring to users' attention.

F: Re-work the evaluator, again, with a per-object-format fixup
D: routine, so as to be able to cope with section offsets "really"
D: being pure numbers; should be able to allow at _least_ the two
D: common idioms
D:   TIMES 510-$ DB 0            ; bootsector
D:   MOV AX,(PROG_END-100H)/16   ; .COM TSR
D: Would need to call the fixup throughout the evaluator, and the
D: fixup would have to be allowed to return UNKNOWN on pass one if it
D: had to. (_Always_ returning UNKNOWN on pass one, though a lovely
D: clean design, breaks the first of the above examples.)
V: 1.10

F: Preprocessor identifier concatenation?
V: 1.10

F: Arbitrary section names in `bin'.
V: 0.98.09
D: Is this necessary? Is it even desirable?
D: hpa: Desirable, yes.  Necessary?  Probably not, but there are definitely cases where it becomes quite useful.
R: madfire
C: 100%

F: Ability to read from a pipe.
V: 1.10
D: Obviously not useful under dos, so memory problems with storing
D: entire input file aren't a problem either.

F: File caching under DOS/32 bit...
V: 1.10?
D: maybe even implement discardable buffers that get thrown away
D: when we get a NULL returned from malloc(). Only really useful under
D: DOS. Think about it.

F: possibly spool out the pre-processed stuff to a file, to avoid having to re-process it.
V: 1.10?
D: Possible problems with preprocessor values not known on pass 1? Have a look...

F: Or maybe we can spool out a pre-parsed version...?
V: 1.10
D: Need to investigate feasibility. Does the results from the parser
D: change from pass 1 to pass 2? Would it be feasible to alter it so that
D: the parser returns an invariant result, and this is then processed
D: afterwards to resolve label references, etc?

F: Subsection support?

F: A good ALIGN mechanism, similar to GAS's.
V: 0.98p1
D: GAS pads out space by means of the following (32-bit) instructions:
D:         8DB42600000000    lea esi,[esi+0x0]
D:         8DB600000000      lea esi,[esi+0x0]
D:         8D742600          lea esi,[esi+0x0]
D:         8D7600            lea esi,[esi+0x0]
D:         8D36              lea esi,[esi]
D:         90                nop
D: It uses up to two of these instructions to do up to 14-byte pads;
D: when more than 14 bytes are needed, it issues a (short) jump to
D: the end of the padded section and then NOPs the rest. Come up with
D: a similar scheme for 16 bit mode, and also come up with a way to
D: use it - internal to the assembler, so that programs using ALIGN
D: don't knock over preprocess-only mode.
D:   Also re-work the macro form so that when given one argument in a
D: code section it calls this feature.
R: Panos Minos
C: 100%?

F: Possibly a means whereby FP constants can be specified as immediate operands to non-FP instructions.
D: * Possible syntax: MOV EAX,FLOAT 1.2 to get a single-precision FP
D:   constant. Then maybe MOV EAX,HI_FLOAT 1.2 and MOV EAX,LO_FLOAT
D:   1.2 to get the two halves of a double-precision one. Best to
D:   ignore extended-precision in case it bites.
D: * Alternatively, maybe MOV EAX,FLOAT(4,0-4,1.2) to get bytes 0-4
D:   (ie 0-3) of a 4-byte constant. Then HI_FLOAT is FLOAT(8,4-8,x)
D:   and LO_FLOAT is FLOAT(8,0-4,x). But this version allows two-byte
D:   chunks, one-byte chunks, even stranger chunks, and pieces of
D:   ten-byte reals to be bandied around as well.

F: A UNION macro might be quite cool
D: now that ABSOLUTE is sane enough to be able to handle it.

F: An equivalent to gcc's ## stringify operator, plus string concatenation
D: somehow implemented without undue ugliness, so as
D: to be able to do `%include "/my/path/%1"' in a macro, or something
D: similar...

F: Actually _do_ something with the processor, privileged and
D: undocumented flags in the instruction table. When this happens,
D: consider allowing PMULHRW to map to either of the Cyrix or AMD
D: versions?
D: hpa: The -p option to ndisasm now uses this to some extent.
V: 1.10

F: Maybe NEC V20/V30 instructions?					?
D: hpa: What are they?  Should be trivial to implement.

F: Yet more object formats.
D: * Possibly direct support for .EXE files?
V: 1.10

F: Symbol map in binary format. Format-specific options...
V: 1.10?

F: REDESIGN: Think about EQU dependency, and about start-point specification in OBJ. Possibly re-think directive support.
V: 1.20?

F: Think about a wrapper program like gcc?
V: 2.00?
D: Possibly invent a _patch_ for gcc so that it can take .asm files on the command line?
D: If a wrapper happens, think about adding an option to cause the
D: resulting executable file to be executed immediately, thus
D: allowing NASM source files to have #!... (probably silly)

F: Multi-platform support?
D: If so: definitely Alpha; possibly Java byte code;
D: probably ARM/StrongARM; maybe Sparc; maybe Mips; maybe
D: Vax. Perhaps Z80 and 6502, just for a laugh?

F: Consider a 'verbose' option that prints information about the resulting object file onto stdout.

F: Line numbers in the .lst file don't match the line numbers in the input.
D: They probably should, rather than the current matching of the post-preprocessor line numbers.