The error was due to casting `const` away for permuting argv
elements, which is necessary for a libc for compatibility with older
systems, but not for us.
Checkpatch will complain about the style not being followed, but this is not
our code, so it can be ignored.
Both `getopt_long` and `getopt_long_only` are GNU-specific, so we'll be
copying musl's implementation for portability.
This was retrieved as of commit 90251cf73dfdd44e7a3f085d236e89a7dff1b00b.
musl is licensed as MIT, which is compatible (being identical...) to RGBDS'.
The file is being copied as-is, without a copyright notice or attribution,
but this is only to have a verbatim copy in the history. Those will be added
in the next commit.
The goal was to improve readability, but along the way a few things were
gained.
- Sorted sym and map files
- Infrastructure for supporting multiple .o versions
- Valgrind-proof, as far as my testing goes anyways
- Improved verbosity messages
- Added error checking
- Performance improvements, see end of commit message
The readability improvement was spurred while trying to make sense of the
old code while trying to implement features such as sorted sym and map
files.
I also did my best to remove hardcoded logic, such that modifications
should be doable; for example, "RAM loading" sections, which are linked
against a different location than the one they're stored at.
Some work remains to be done, see the "TODO:" and "FIXME:" comments.
Further, while regression tests pass, this new linker should be tested on
different codebases (ideally while instrumented with `make develop` and
under valgrind).
The few errors spotted in the man pages (alignment) need to be corrected.
Finally, documentation comments need to be written, I have written a lot of
them but not all.
This also provides a significant performance boost (benchmarked with a
51994-symbol project):
Current master RGBLINK:
2.02user 0.03system 0:02.06elapsed 99%CPU (0avgtext+0avgdata 84336maxresident)k
0inputs+11584outputs (0major+20729minor)pagefaults 0swaps
Rewritten RGBLINK:
0.19user 0.06system 0:00.63elapsed 40%CPU (0avgtext+0avgdata 32460maxresident)k
23784inputs+11576outputs (0major+7672minor)pagefaults 0swaps
This is especially useful when an EQUS expands to another one, to help
track them.
This is done separately from the file stack as the EQUS stack is separate
(which is itself because EQUS are managed *way* differently).
Some errors are only tripped in `out_WriteObject`, which was
basically a stub when `-o` wasn't specified. Now, instead,
errors are checked in a separate function before out_WriteFile
The old error stack was fairly obtuse and hard to use for debugging.
This improves it notably by ensuring all line numbers are relative
to the file, and not, say, the macro definition.
This is a breaking change if you were parsing the old stack, but
the change should be painless, and the new stack only brings more info.
The syntax is unchanged for files, macros see their name prefixed
with the file they're defined in and a pair of colors, REPT blocks
simply append a '::REPT~n' to the context they're in, where 'n' is
the number of iterations the REPT has done.
This is especially helpful in macro-heavy code such as rgbds-structs.
Unlike macros, REPTs and INCLUDEs, this recursion depth is independent.
This is intentional, because string expansions work very differently.
While it's easy to know when a string expansion begins, checking where it
ends is much more complicated, since the expansion's contents are simply
injected back into the lex buffer. Therefore, the depth has to be checked
after lexing took place.
Because of this, the placement of the expansion end check is somewhat
haphazard, but I think it's good. While I have no certainty, all tests
ended with all expansions properly ended, and I couldn't find any pitfalls.
Finally, `pCurrentStringExpansion` has been made global so error printing
can use it to tell the user if an error occurred inside of an expansion.
REPT blocks nested in macros (and possibly other cases) leaked
memory on every call. Unlike most other memory leaks, which would
be freed at the end of program execution if they were done properly,
those piled up the more compilation went on.
I believe memory usage could have started being fairly high on
large projects following the "one master file INCLUDEs all the rest"
so this may have actually been worth it.
(And REPT.)
Not exactly a *recursion* limit, more like a *stack depth* limit,
but calling it "recursion" conveys its purpose better.
The default of 64 is super overkill: even in a a project with
what I believe to be above-average levels of nesting, the
level only peaked at 6.
Keeping in mind the purpose of this is to catch infinite
recursion, which is still caught quickly (in usual cases, anyways),
this default seems sensible.
And it passes tests. What more do you need?
This adds two new directives: newcharmap and setcharmap.
newcharmap creates a new charmap and switches to it.
setcharmap switches to an existing charmap.
Should partially cover #178 and close#270.
This allows printing numbers in different bases and without the dollar prefix
This is especially useful in macros because the dollar isnt a valid character
for symbol names, requiring heavy `STRSUB` usage.
Charmap's previous structure was using brute-force comparison for
converting the strings in source files. It always compared given
strings to all of the strings in charmap, which was very costly
in huge projects.
For its improvement, I changed its structure into trie, which is
being used in many string-processing areas. It's now much faster
than before.
The createpatch() function was using a fixed-size buffer. I've changed it
to be dynamically allocated. I saw that the RPN format used in patches is
slightly different from the one used internally in the assembler, so I
added a new member to the Expression struct to track the patch size.
I've also limited the RPN expression length to 1MB. I realized that the
patch RPN expression could potentially be longer than the internal RPN
expression, so the internal expression would need a limit smaller than
UINT32_MAX. I thought 1MB would be a reasonable limit.
The linker script now allows you to assign a section with the same
attributes as in the source.
To do this, I've removed a check from AssignSectionAddressAndBankByName
that would never be triggered, due to that condition being checked
before. Shouldn't this and IsSectionSameTypeBankAndAttrs be condensed
into a single function?
