this is a bit complicated: for i386 and x86-64 we really need to
extend return values ourself, as the common code now does. For arm64
this at least preserves old behaviour. For riscv64 we don't have to
extend ourself but can expect things to be extended up to int (this
matters for var-args tests, when the sign-extension to int64 needs to
happen explicitely). As the extensions are useless, don't do them.
And for arm32 we actually can't express GCC behaviour: the callee side
expects the return value to be correctly extended to int32, but
remembers the original type. In case the ultimate target type for the
call result is only int, no further extension is done. But in case
the target type is e.g. int64 an extension happens, but not from int32
but from the original type. We don't know the ultimate target type,
so we have to choose a type to put into vtop:
* original type (plus VT_MUSTCAST) - this looses when the ultimate
target is int (GCC: no cast, TCC: a cast)
* int (without MUSTCAST) - this looses when the ultimate target is
int64 (GCC: cast from original type, TCC: cast from int)
This difference can only be seen with undefined sources, like the
testcases, so it doesn't seem worthwhile to try an make it work, just
disable the test on arm and choose the second variant as that generates
less code.
This allows adding files or libraries from
#pragma comment(option, ...)
Also, {f}/file.c will be expanded with the directory of
the current source, that is the file that has the #pragma
Put total_lines etc. into TCCState. Also, initialize
the predefined compiler types for the preprocessor too.
tccpe.c: fix BaseOfCode if .init section present (with tcc -b)
protect some more accesses to global data with the semaphore.
(And for the testcase: don't just write into global data, use local
copies; it's not important for speed here).
* a major revision of the rt_printline() feature in
tccrun.c to report file:linenumber more correctly.
* minor changes to the stab info produced by the
compiler in tccgen.c
However stab addresses are limited to 32 bits. I added
a work around:
if (sizeof pc == 8)
pc |= wanted_pc & 0xffffffff00000000ULL;
However GDB has problems with that too.
- revert Makefiles to state before last bcheck additions
Instead, just load bcheck.o explicitly if that is
what is wanted.
- move tcc_add_bcheck() to the <target>-link.c files and
remove revently added arguments. This function is to
support tccelf.c with linking, not for tccgen.c to
support compilation.
- remove -ba option: It said:
"-ba Enable better address checking with bounds checker"
Okay, if it is better then to have it is not an option.
- remove va_copy. It is C99 and we try to stay C89 in tinycc
when possible. For example, MS compilers do not have va_copy.
- win64: revert any 'fixes' to alloca
It was correct as it was before, except for bound_checking
where it was not implemented. This should now work too.
- remove parasitic filename:linenum features
Such feature is already present with rt_printline in
tccrun.c. If it doesn't work it can be fixed.
- revert changes to gen_bounded_ptr_add()
gen_bounded_ptr_add() was working as it should before
(mostly). For the sake of simplicity I switched it to
CDECL. Anyway, FASTCALL means SLOWCALL with tinycc.
In exchange you get one addition which is required for
bounds_cnecking function arguments. The important thing
is to check them *BEFORE* they are loaded into registers.
New function gbound_args() does that.
In any case, code instrumentation with the bounds-check
functions as such now seems to work flawlessly again,
which means when they are inserted as NOPs, any code that
tcc can compile, seems to behave just the same as without
them.
What these functions then do when fully enabled, is a
differnt story. I did not touch this.
There was a problem with strncpy and strncmp.
Made bound_ptr_add and bound_ptr_indir a little bit faster.
Fix statistic counter types. Change long into long long.
The following functions are now also bounds checked:
memcmp, strncpy, strcmp, strncmp, strcat, strchr, strdup.
Add statistics code for bounds checking functions.
The statistics can be printed by settings environment variable
"TCC_BOUNDS_PRINT_STATISTIC".
Enabled more tests in test/Makefile.
The bounds checking code has now enabled gen_bounded_ptr_add tests.
This makes the code slower but finds more errors.
I had to correct some things in tcc to make it work.
- Fixed off by one in lib/bcheck.c
- Corrected tccelf.c sym_versions.
- Disabled USE_TAL when using bounds checking.
- Fixed cstr_printf va_start.
- Fixed tests/tests2/46_grep.c off by one error.
- Updated gen_bounded_ptr_add in x86_64-gen.c
- Fixed x86_64-link.c pointer diff.
For gen_vla_alloc now always use alloca call when bounds checking.
Added line/filename in %rax before bound calls to find location of error.
my last patch here fixed a missing dependency, but in the wrong
way. It would build some object files twice, which could race.
The lib/all target already builds both libtcc{,b}1.a libs, so
a simple dependency is enough.
This allows creation of TCCStates and operation with API
calls independently from each other, even from threads.
Frontend (option parsing/libtcc.c) and backend (linker/tccelf.c)
now depend only on the TCCState (s1) argument.
Compilation per se (tccpp.c, tccgen.c) is still using
globals for convenience. There is only one entry point
to this section which is tcc_compile() which is protected
by a semaphore.
There are some hacks involved to avoid too many changes,
as well as some changes in order to avoid too many hacks ;)
The test libtcc_test_mt.c shows the feature. Except this
new file the patch adds 87 lines overall.
on i386 111_conversion.c breaks when save_reg_upstack isn't careful
about r2 and type mismatches. The bcheck patches fixed this by
enlarging the stack slot beyond the natural type, this variant simply
avoids saving the second register is the type indicates that it isn't
needed.
Adds also a comment how this should ideally work, namely that type
and r/r2 entries in the vstack are consistent. In the 111_conversion
testcase it's specifically gen_cast via gen_cvt_ftoi that breaks
this, but there more general code broken as well, so that would deserve
a careful fixup based on some additional asserts.
so it isn't quadratic in number of symbols. It's still quadratic
in number of lib/version combinations (at library load time the
sym versions are internalized), but that's much less a problem as
in practice only glibc uses sym versioning.
