The parser uses "break" as fake label to compile "break" as "goto
break". To avoid producing this string at each use, it keeps it
available in its state.
Added comments so that all braces pair correctly. (The parser has
several instances of unmatched braces as characters ('{' or '}'), which
hinders matching regular braces in the code.)
The compilation of a goto or a label just create an entry and generate
boilerplate code for the gotos. As we don't know yet whether it needs a
CLOSE, we code a jump followed by a CLOSE, which is then dead code.
When a block ends (and then we know for sure whether there are variables
that need to be closed), we check the goto's against the labels of that
block. When closing a goto against a label, if it needs a CLOSE, the
compiler swaps the order of the jump and the CLOSE, making the CLOSE
active.
Moreover, each function being parsed has its own table.
The code is cleaner when each table is used for one specific purpose:
The scanner uses its table to anchor and unify strings, mapping strings
to themselves; the parser uses it to reuse constants in the code,
mapping constants to their indices in the constant table. A different
table for each task avoids false collisions.
New instruction format 'ivABC' (a variant of iABC where parameter vC has
10 bits) allows constructors of up to 1024 elements to be coded without
EXTRAARG.
Instead of a fixed limit of 50 registers (which, in a bad worst case,
can limit the nesting of constructors to 5 levels), the compiler
computes an individual limit for each constructor based on how many
registers are available when it runs. This limit then controls the
frequency of SETLIST instructions.
Several definitions that don't need to be "global" (that is, that
concerns only specific parts of the code) moved out of llimits.h,
to more appropriate places.
Function 'leaveblock' was generating "break" label before removing
variables from the closing block. If 'createlabel' created a 'close'
instruction (which it did when matching a goto/break that exited
the scope of an upvalue), that instruction would use the wrong level.
More uses of macros 'likely'/'unlikely' (renamed to
'l_likely'/'l_unlikely'), both in range (extended to the
libraries) and in scope (extended to hooks, stack growth).
A "with stack" implementation gains too little in performance to be
worth all the noise from C-stack overflows.
This commit is almost a sketch, to test performance. There are several
pending stuff:
- review control of C-stack overflow and error messages;
- what to do with setcstacklimit;
- review comments;
- review unroll of Lua calls.
The parser were mixing compiler indices of variables with stack indices,
so that when a to-be-closed variable was used inside the scope of
compile-time constants (which may be optimized away), it might be closed
in the wrong place. (See new tests for examples.)
Besides fixing the bugs, this commit also changed comments and variable
names to avoid that kind of confusion and added tests.
The code should not compute an instruction address before checking that
it exists. (Virtually no machine complains of computing an invalid
address, as long as the address is not used, but for ISO C that is
undefined behavior.)
This function was computing invalid instruction addresses when the
expression was not a multi-return instruction. (Virtually all machines
don't raise errors when computing an invalid address, as long as the
address is not accessed, but this computation is undefined behavior in
ISO C.)
Attributes changed to posfixed ('x <const>', instead of '<const> x'),
and "toclose" renamed to "close". Posfixed attributes seem to make it
clearer that it applies to only one variable when there are multiple
variables.
