`&nil` is recently introduced in CRuby to explicitly declare that
a method does not accept a block. When a block is passed,
ArgumentError "no block accepted" is raised. This is analogous to
`**nil` for keyword arguments.
The noblock flag is encoded in bit 23 of OP_ENTER's aspec operand
(24=n1:m5:o5:r1:m5:k5:d1:b1), avoiding the need for a new opcode.
Co-authored-by: Claude <noreply@anthropic.com>
These opcodes use BB format instead of BBB, saving 1 byte per call.
In the standard library, this saves ~790 bytes (568 SEND0 + 222 SSEND0).
Co-authored-by: Claude <noreply@anthropic.com>
Add single-byte opcodes for returning true/false directly, completing
the set of literal return opcodes (RETSELF, RETNIL, RETTRUE, RETFALSE).
Codegen applies peephole optimization to fuse LOADTRUE/LOADFALSE + RETURN.
Co-authored-by: Claude <noreply@anthropic.com>
Rename boolean load opcodes for consistency with LOADNIL/LOADSELF.
Backward compatibility aliases are provided in opcode.h.
Co-authored-by: Claude <noreply@anthropic.com>
Add a new opcode that returns nil without requiring LOADNIL + RETURN.
This avoids loading nil into a register by setting the return value (v)
directly. The implementation uses a separate label (L_RETURN_NIL) to
bypass v = regs[a], preserving self in regs[0] for ensure blocks.
Codegen applies peephole optimization to fuse LOADNIL + RETURN -> RETNIL.
Co-authored-by: Claude <noreply@anthropic.com>
Bypass method dispatch when calling blocks via yield. The new OP_BLKCALL
instruction directly invokes the proc without looking up Proc#call,
resulting in 13-17% faster yield performance.
Co-authored-by: Claude <noreply@anthropic.com>
TDEF fuses TCLASS+METHOD+DEF for normal method definitions.
SDEF fuses SCLASS+METHOD+DEF for singleton method definitions.
Saves 4 bytes per method definition (8 bytes -> 4 bytes).
Falls back to unfused instructions if irep index exceeds 255.
Co-authored-by: Claude <noreply@anthropic.com>
Fuses MOVE+LOADI_0+GETIDX pattern into single instruction.
Saves 4 bytes per arr[0] access (7 bytes -> 3 bytes).
Co-authored-by: Claude <noreply@anthropic.com>
fuse MOVE+ADDI+MOVE and MOVE+SUBI+MOVE patterns into single instructions.
ADDILV/SUBILV add/subtract an immediate to a local variable in-place.
BBB format: a=local, b=working space for method call, c=immediate.
saves 5 bytes per instance (9->4 bytes), 40 occurrences in stdlib.
Co-authored-by: Claude <noreply@anthropic.com>
Change OP_MATCHERR from Z format (unconditional) to B format
(conditional on register). This allows fusing JMPIF + MATCHERR
sequence into a single MATCHERR instruction for simple patterns.
Before: JMPIF R2 target (4 bytes) + MATCHERR (1 byte) = 5 bytes
After: MATCHERR R2 (2 bytes)
Saves 3 bytes per pattern match with raise_on_fail.
Co-authored-by: Claude <noreply@anthropic.com>
Replace 4-instruction sequence (GETCONST + STRING + SEND + RAISEIF)
with single OP_MATCHERR instruction that raises NoMatchingPatternError
with "pattern not matched" message.
Bump RITE binary format version from 0300 to 0400 due to opcode
number shift.
Co-authored-by: Claude <noreply@anthropic.com>
OP_LOADI stores an 8 bit integer to a register, so we renamed the
instruction name to describe the behavior more precisely, like
OP_LOADI16 and OP_LOADI32.
Transforms the value of a splat inside a return statement (similar
to an array). For example, `return *nil` should return `nil.to_a`,
while `return *1` should return `[1]`
The Difference
Since Ruby1.9, the keyword arguments were emulated by Ruby using the hash
object at the bottom of the arguments. But we have gradually moved toward
keyword arguments separated from normal (positinal) arguments.
At the same time, we value compatibility, so that Ruby3.0 keyword
arguments are somewhat compromise. Basically, keyword arguments are
separated from positional arguments, except when the method does not
take any formal keyword arguments, given keyword arguments (packed
in the hash object) are considered as the last argument.
And we also allow non symbol keys in the keyword arguments. In that
case, those keys are just passed in the `**` hash (or raise
`ArgumentError` for unknown keys).
The Instruction Changes
We have changed `OP_SEND` instruction. `OP_SEND` instruction used to
take 3 operands, the register, the symbol, the number of (positional)
arguments. The meaning of the third operand has been changed. It is now
considered as `n|(nk<<4)`, where `n` is the number of positional
arguments, and `nk` is the number of keyword arguments, both occupies
4 bits in the operand.
The number `15` in both `n` and `nk` means variable sized arguments are
packed in the object. Positional arguments will be packed in the array,
and keyword arguments will be packed in the hash object. That means
arguments more than 14 values are always packed in the object.
Arguments information for other instructions (`OP_SENDB` and `OP_SUPER`)
are also changed. It works as the third operand of `OP_SEND`. the
difference between `OP_SEND` and `OP_SENDB` is just trivial. It assigns
`nil` to the block hidden arguments (right after arguments).
The instruction `OP_SENDV` and `OP_SENDVB` are removed. Those
instructions are replaced by `OP_SEND` and `OP_SENDB` respectively with
the `15` (variable sized) argument information.
Calling Convention
When calling a method, the stack elements shall be in the order of the
receiver of the method, positional arguments, keyword arguments and the
block argument. If the number of positional or keyword arugument (`n` or
`nk`) is zero, corresponding arguments will be empty. So when `n=0` and
`nk=0` the stack layout (from bottom to top) will be:
+-----------------------+
| recv | block (or nil) |
+-----------------------+
The last elements `block` should be explicitly filled before `OP_SEND`
or assigned to `nil` by `OP_SENDB` internally. In other words, the
following have exactly same behavior:
OP_SENDB clears `block` implicitly:
```
OP_SENDB reg sym 0
```
OP_SEND clears `block` implicitly:
```
OP_LOADNIL R2
OP_SEND R2 sym 0
```
When calling a method with only positional arguments (n=0..14) without
keyword arguments, the stack layout will be like following:
+--------------------------------------------+
| recv | arg1 | ... | arg_n | block (or nil) |
+--------------------------------------------+
When calling a method with arguments packed in the array (n=15) which
means argument splat (*) is used in the actual arguments, or more than
14 arguments are passed the stack layout will be like following:
+-------------------------------+
| recv | array | block (or nil) |
+-------------------------------+
The number of the actual arguments is determined by the length of the
argument array.
When keyword arguments are given (nk>0), keyword arguments are passed
between positional arguments and the block argument. For example, when
we pass one positional argument `1` and one keyword argument `a: 2`,
the stack layout will be like:
+------------------------------------+
| recv | 1 | :a | 2 | block (or nil) |
+------------------------------------+
Note that keyword arguments consume `2*nk` elements in the stack when
`nk=0..14` (unpacked).
When calling a method with keyword arguments packed in the hash object
(nk=15) which means keyword argument splat (**) is used or more than
14 keyword arguments in the actual arguments, the stack layout will
be like:
+------------------------------+
| recv | hash | block (or nil) |
+------------------------------+
Note for mruby/c
When mruby/c authors try to support new keyword arguments, they need
to handle the new meaning of the argument information operand. If they
choose not to support keyword arguments in mruby/c, it just raise
error when `nk` (taken by `(c>>4)&0xf`) is not zero. And combine
`OP_SENDV` behavior with `OP_SEND` when `n` is `15`.
If they want to support keyword arguments seriously, contact me at
<matz@ruby.or.jp> or `@yukihiro_matz`. I can help you.
Which represent `obj[int]` and `obj[int]=val` respectively where `obj`
is either `string`, `array` or `hash`, so that index access could be
faster. When `obj` is not assumed type or `R(a+1)` is not integer, the
instructions fallback to method calls.
Add n elements at once. Reduces instructions for huge array
initialization. In addition, `gen_value` function in `codegen.c` was
refactored and clarified.
This reverts commit fd10c72319.
I thought it was OK to restrict index value within 1 byte, but in some
cases index value could be 16 bits (2 bytes). I had several ideas to
address the issue, but reverting `fd10c72` is the easiest way. The
biggest reason is `mruby/c` still supports `OP_EXT[123]`, so that they
don't need any additional work.
We have introduced following new instructions.
* `OP_LAMBDA16`
* `OP_BLOCK16`
* `OP_METHOD16`
* `OP_EXEC16`
Each instruction uses 16 bits operand for `reps` index. Since new
instructions are added, `mruby/c` VM should be updated.
Due to new instructions, dump format compatibility is lost, we have
increased `RITE_BINARY_MAJOR_VER`.
In addition, we have decreased the size of `refcnt` in `mrb_irep` from
`uint32_t` to `uint16_t`, which is reasonably big enough.
Jump target address is `operand (16bit)` + `address of next instruction`.
In addition, `ilen` was made `uint32_t` so that `iseq` length limitation
of 65536 is removed. Only jump target address should be within signed
16bit (-32768 .. 32767).
New instructions:
* OP_LOADL16
* OP_LOADSYM16
* OP_STRING16
Size of pools, symbols are `int16_t` but offset representation in the
bytecode was 8 bits. Size of child `irep` array is `int16_t`, too but
this change does not address it.
When a global jump occurs, look at the catch handler table to determine where to jump.
In that case, `pc` already shows the following instruction, but since the table shows `begin_offset ... end_offset`, the comparison is done with `begin_offset < pc && pc <= end_offset`.
If there is a corresponding handler, move `pc` to `handler.target_offset` and continue running the VM.
When a global jump across `ensure` is made by `return`, `break`, `next`, `redo` and `retry`, the extended `RBreak` object saves and restores the C-level execution position.
This extended `RBreak` can have tag information, which makes it a pseudo coroutine (the "tag" mimics CRuby).
The implementation of pseudo coroutines by `RBreak` is summarized by `CHECKPOINT_RESTORE ... CHECKPOINT_MAIN ... CHECKPOINT_END` and `throw_tagged_break` / `unwind_ensure` macros.
The restart of processing is branched by `RBREAK_TAG_FOREACH(DISPATCH_CHECKPOINTS)`.
- Not only `rescue` blocks but also `ensure` blocks are now sandwiched between `OP_EXCEPT` and `OP_RAISEIF`.
- Remove the function `ecall()`.
It is no longer necessary to re-enter the VM to perform an "ensure block".
This will resolves#1888.
- Added instruction `OP_JUW` (Jump while UnWind).
It jumps unconditionally like `OP_JMP`, but searches the catch handler table and executes the ensure block.
Since it searches the catch handler table, it is much heavier than `OP_JMP`.
- `OP_EXCEPT` checks if `mrb->exc` is `NULL`, `MRB_TT_EXCEPTION` or
`MRB_TT_BREAK`.
If `mrb->exc` is `NULL`, it will be replaced with `nil`.
- If `OP_RAISE` is `nil`, it does nothing and the immediately
following instruction is executed (like `OP_NOP`).
Also, in case of `RBreak` object, it moves to the processing for
`break`.
With this change, the instruction name is changed from
`OP_RAISE` to `OP_RAISEIF`.
Which loads 16bit integer to the register. The instruction number should
be reorder on massive instruction refactoring. The instruction is added
for `mruby/c` which had performance issue with `OP_EXT`. With this
instruction, `mruby/c` VM can just raise errors on `OP_EXT` extension
instructions.
Yukihiro "Matz" Matsumoto