| Age | Commit message (Collapse) | Author |
|---|
|
Because we can't use the highest 2 bits of function pointers.
|
|
Because we can't use the highest 2 bits of function pointers.
|
|
|
|
I think they can always be enabled because the regular expression literal is
always enabled.
|
|
linked programs
In lld linked programs, .rodata comes before .text, thus mrb_ro_data_p
will return false for strings in .rodata. Change the lower bound from
_etext to __ehdr_start to catch these cases. This works for ld.bfd, gold
and lld, and it does not have false positives even if .init_array does
not exist.
Remove the branch that uses _edata: strings in .data can be modified so
this is semantically incorrect. Delete the __APPLE__ branch (its
manpages say get_etext() and get_edata() are strongly discouraged).
.init_array has been adopted by most ELF platforms to supersede .ctors.
Neither _etext nor _edata is used, so rename MRB_USE_ETEXT_EDATA to
MRB_USE_EHDR_START.
|
|
`MRB_METHOD_TABLE_INLINE` was fragile. It requires `-falign-functions=n`.
On platform that uses higher bits of function pointers, you can use new
`MRB_METHOD_T_STRUCT` configuration macro.
|
|
Not only mruby, it is one of the difficult things to adjust the performance vs. memory efficiency of the software.
If the approximate target device is divided and the adjustment value for it is prepared by default, it is a good indicator to do fine adjustment.
This PR divides into four profiles.
***Caution: There is no basis for the definitions in the patch.***
- `MRB_CONSTRAINED_BASELINE_PROFILE` - for microprocessors. Reduce memory consumption.
- `MRB_BASELINE_PROFILE` - Default value of mruby.
- `MRB_MAIN_PROFILE` - For desktop computers. Assume that a huge amount of RAM, 10 MiB or more, is on board.
- `MRB_HIGH_PROFILE` - for servers. Assume that mruby VM has a long life.
As you can see the profile name has been ~~stolen~~ imitated from H.264.
|
|
|
|
|
|
- (Modify) `MRB_INT16`
- (Add) `MRB_INT32`
- (Modify) `MRB_INT64`
- (Add) `MRB_USE_ETEXT_EDATA`
- (Add) `MRB_NO_INIT_ARRAY_START
- (Add) `MRB_WITHOUT_FLOAT`
- (Add) `MRB_METHOD_CACHE`
- (Add) `MRB_METHOD_CACHE_SIZE`
- (Add) `MRB_METHOD_TABLE_INLINE
- (Add) `MRB_ENABLE_ALL_SYMBOLS`
|
|
|
|
|
|
`cpp` does not raise error on undefined macro access in condition.
|
|
You had to define this macro on big endian platforms, but it is very
error-prone. So define the macro automatically if possible.
|
|
It was default on Linux. Abandoned for 2 reasons:
* some cross-platform compiler defines `__linux__` even on bare metal
environment (RTL8196 (MIPS big endian soc) for example).
* some compilers do not align functions pointers so that we need to
specify `-falign-functions=n` (where n>1). It's not suitable for
default configuration.
By our measurement, `mrbtest` consumes 400KB less memory. So if your
target machine is memory tight, add the following command-line options
to `GCC` (unfortunately `clang` does not support `-falign-functions`).
`-falign-functions=2 -DMRB_METHOD_TABLE_INLINE`
|
|
Now the method tables (in classes/modules and caches) keeps C function
pointers without wrapping in `struct RProc` objects. For the sake of
portability, `mrb_method_t` is represented by the struct and union, but
if the most significant bit of the pointer is not used by the platform,
`mrb_method_t` should be packed in `uintptr_t` to reduce memory usage.
`MRB_METHOD_TABLE_INLINE` is turned on by default for linux.
|
|
|
|
|
|
|
|
Unless `MRB_NAN_BOXING` is specified.
|
|
|
|
|
|
|
|
|
|
|
|
changes:
* rename DISABLE_STDIO -> MRB_DISABLE_STDIO
* rename ENABLE_DEBUG -> MRB_ENABLE_DEBUG_HOOK
* no more opposite macro definitions (e.g. ENABLE_STDIO, DISABLE_DEBUG).
* rewrite above macro references throughout the code.
* update documents
|
|
define MRB_UTF8_STRING (in mrbconf.h) to enable UTF-8 support.
|
|
|
|
b72e94f used _etext and _edata to distinguish C string literals from heap allocated strings,
but using _edata, global char arrays may be considered as string literals. to avoid the issue,
we have to use __init_array_start, which might be less portable. you can still use _edata, by
using MRB_NO_INIT_ARRAY_START.
|
|
on platforms with _etext and _edata, mruby can distinguish string literals so that it avoids memory allocation to copy them.
for example, on my Linux box (x86 32bit), memory consumed by mrbtest decreased from 8,168,203 to 8,078,848 (reduced 88KB).
|
|
|
|
Adds following macros:
* MRB_FIXED_STATE_ATEXIT_STACK (not defined by default)
* When defined enables fixed state atexit stack.
* MRB_FIXED_STATE_ATEXIT_STACK_SIZE (default value: 5)
* This macro will be ignored when `MRB_FIXED_STATE_ATEXIT_STACK` isn't defined.
* When `mrb_state_atexit` is called more than this value it will raise runtime error.
|
|
|
|
|
|
According to CONTRIBUTING.md,
Don't use C++ style comments
/* This is the prefered comment style */
Use C++ style comments only for temporary comment e.g. commenting out some code lines.
|
|
|
|
|
|
You will not see "arena overflow" error anymore, but I encourage gem authors
to check your gems with MRB_GC_FIXED_ARENA to avoid memory broat.
|
|
MRB_GC_TURN_OFF_GENERATIONAL; #1447
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
It is not used in the core.
It does not have mrb_ prefix.
strtol() is slightly heavy and we have similar API.
|
|
|
|
stddef.h is included indirectly.
|
|
|
