To prevent cache misses, token ids go in their own array, and the
start/end offsets go in a different one.
perf measurement before:
2,667,914 cache-misses:u
2,139,139,935 instructions:u
894,167,331 cycles:u
perf measurement after:
1,757,723 cache-misses:u
2,069,932,298 instructions:u
858,105,570 cycles:u
Now there are 3 types:
* std.math.big.int.Const
- the memory is immutable, only stores limbs and is_positive
- all methods operating on constant data go here
* std.math.big.int.Mutable
- the memory is mutable, stores capacity in addition to limbs and
is_positive
- methods here have some Mutable parameters and some Const
parameters. These methods expect callers to pre-calculate the
amount of resources required, and asserts that the resources are
available.
* std.math.big.int.Managed
- the memory is mutable and additionally stores an allocator.
- methods here perform the resource calculations for the programmer.
- this is the high level abstraction from before
Each of these 3 types can be converted to the other ones.
You can see the use case for this in the self-hosted compiler, where we
only store limbs, and construct the big ints as needed.
This gets rid of the hack where the allocator was optional and the
notion of "fixed" versions of the struct. Such things are now modeled
with the `big.int.Const` type.
Pre-requisite for having a test case for #5062
In complex C statements which are outside of macros,
it is valid C to perform e.g. a bitor between an
integer and a boolean `5 | (8 == 9)`
Currently this results in a zig error after translating
as `c_int | bool` is invalid Zig.
Detects if a sub-expression of a numeric operator is
boolean and if so converts it to int
See https://github.com/ifreund/river/issues/17 for an issue that occurs
because the field names are mangled globally. When using the generated
bindings, you have no choice but to use the unstable names or redeclare
the entire struct. This commit changes the behaviour to use a local
counter per record declaration, so the names are predictable each time.
