* anonymous decls have automatically generated names and symbols, and
participate in the same memory management as named decls.
* the Ref instruction is deleted
* the DeclRef instruction now takes a `[]const u8` and DeclRefStr takes
an arbitrary string instruction operand.
* introduce a `zir.Decl` type for ZIR Module decls which holds
content_hash and name - fields that are not needed for `zir.Inst`
which are created as part of semantic analysis. This improves the
function signatures of Module.zig and lowers memory usage.
* the Str instruction is now defined to create an anonymous Decl and
reference it.
zig fmt previously would write a temp file, and then either rename it
into place if necessary, or unlink it if nothing was changed. Now zig
fmt renders into a memory buffer, and only writes the temp file and
renames it into place if anything changed.
Based on the performance testing I did this actually did not have much
of an impact, however it's likely that on other operating systems and
other hard drives this could make a big difference.
* `std.fs.Dir.Entry.Kind` is moved to `std.fs.File.Kind`
* `std.fs.File.Stat` gains the `kind` field, so performing a stat() on
a File now tells what kind of file it is. On Windows this only will
distinguish between directories and files.
* rework zig fmt logic so that in the case of opening a file and
discovering it to be a directory, it closes the file descriptor
before re-opening it with O_DIRECTORY, using fewer simultaneous open
file descriptors when walking a directory tree.
* rework zig fmt logic so that it pays attention to the kind of
directory entries, and when it sees a sub-directory it attempts to
open it as a directory rather than a file, reducing the number of
open() syscalls when walking a directory tree.
The original check for a directory was for the `readAllAlloc` so move the check from open to read. This in turn fixes the fmt step in the build script for directories.
* Take advantage of coercing anonymous struct literals to struct types.
* Reworks Module to favor Zig source as the primary use case.
Breaks ZIR compilation, which will have to be restored in a future commit.
* Decl uses src_index rather then src, pointing to an AST Decl node
index, or ZIR Module Decl index, rather than a byte offset.
* ZIR instructions have an `analyzed_inst` field instead of Module
having a hash table.
* Module.Fn loses the `fn_type` field since it is redundant with
its `owner_decl` `TypedValue` type.
* Implement Type and Value copying. A ZIR Const instruction's TypedValue
is copied to the Decl arena during analysis, which allows freeing the
ZIR text instructions post-analysis.
* Don't flush the ELF file if there are compilation errors.
* Function return types allow arbitrarily complex expressions.
* AST->ZIR for function calls and return statements.
* Introduce the concept of anonymous Decls
* Primitive Hello, World with inline asm works
* There is still an unsolved problem of how to manage ZIR instructions
memory when generating from AST. Currently it leaks.
* Deleted decls are deleted; unused decls are also detected as deleted.
Cycles are not yet detected.
* Re-analysis is smarter and will not cause a re-analysis of dependants
when only a function body is changed.
The binary file abstraction changed its struct named "Decl" to
"TextBlock" and it now represents an allocated slice of memory in
the .text section. It has two new fields: prev and next, making it
a linked list node. This allows a TextBlock to find its neighbors.
The ElfFile struct now has free_list and last_text_block fields.
Doc comments for free_list are reproduced here:
A list of text blocks that have surplus capacity. This list can have false
positives, as functions grow and shrink over time, only sometimes being added
or removed from the freelist.
A text block has surplus capacity when its overcapacity value is greater than
minimum_text_block_size * alloc_num / alloc_den. That is, when it has so
much extra capacity, that we could fit a small new symbol in it, itself with
ideal_capacity or more.
Ideal capacity is defined by size * alloc_num / alloc_den.
Overcapacity is measured by actual_capacity - ideal_capacity. Note that
overcapacity can be negative. A simple way to have negative overcapacity is to
allocate a fresh text block, which will have ideal capacity, and then grow it
by 1 byte. It will then have -1 overcapacity.
The last_text_block keeps track of the end of the .text section.
Allocation, freeing, and resizing decls are all now more sophisticated,
and participate in the virtual address allocation scheme. There is no
longer the possibility for virtual address collisions.
