Use std.heap.FixedBufferAllocator combined with
std.heap.DirectAllocator instead.
std.mem.FixedBufferAllocator is moved to std.heap.FixedBufferAllocator
* DirectAllocator does the underlying syscall for every allocation.
* ArenaAllocator takes another allocator as an argument and
allocates bytes up front, falling back to DirectAllocator with
increasingly large allocation sizes, to avoid calling it too often.
Then the entire arena can be freed at once.
The self hosted parser is updated to take advantage of ArenaAllocator
for the AST that it returns. This significantly reduces the complexity
of cleanup code.
docgen and build runner are updated to use the combination of
ArenaAllocator and DirectAllocator instead of IncrementingAllocator,
which is now deprecated in favor of FixedBufferAllocator combined
with DirectAllocator.
The C allocator calls aligned_alloc instead of malloc, in order to
respect the alignment parameter.
Added asserts in Allocator to ensure that implementors of the
interface return slices of the correct size.
Fixed a bug in Allocator when you call realloc to grow the allocation.
* move std.io.File to std.os.File
* add `zig fmt` to self hosted compiler
* introduce std.io.BufferedAtomicFile API
* introduce std.os.AtomicFile API
* add `std.os.default_file_mode`
* change FileMode on posix from being a usize to a u32
* add std.os.File.mode to return mode of an open file
* std.os.copyFile copies the mode from the source file instead of
using the default file mode for the dest file
* move `std.os.line_sep` to `std.cstr.line_sep`
Before we accepted a nullable allocator for some stuff like
opening files. Now we require an allocator.
Use the mem.FixedBufferAllocator pattern if a bound on the amount
to allocate is known.
This also establishes the pattern that usually an allocator is the
first argument to a function (possibly after "self").
fix docs for std.cstr.addNullByte
self hosted compiler:
* only build docs when explicitly asked to
* clean up main
* stub out zig fmt
at - Get the item at the n-th index.
insert - Insert and item into the middle of the list, resizing and copying
existing elements if needed.
insertSlice - Insert a slice into the middle of the list, resizing and
copying existing elements if needed.
Add fallback paths for when the getrandom(2) system call is not
available. Try /dev/urandom first and sysctl(RANDOM_UUID) second.
The sysctl issues a warning in the system logs with some kernels but
that seems like an acceptable tradeoff for the fallback of a fallback.
The purpose of this is:
* Only one way to do things
* Changing a function with void return type to return a possible
error becomes a 1 character change, subtly encouraging
people to use errors.
See #632
Here are some imperfect sed commands for performing this update:
remove arrow:
```
sed -i 's/\(\bfn\b.*\)-> /\1/g' $(find . -name "*.zig")
```
add void:
```
sed -i 's/\(\bfn\b.*\))\s*{/\1) void {/g' $(find ../ -name "*.zig")
```
Some cleanup may be necessary, but this should do the bulk of the work.
* docgen supports obj_err code kind for demonstrating
errors without explicit test cases
* add documentation for `extern enum`. See #367
* remove coldcc keyword and add @setIsCold. See #661
* add compile errors for non-extern struct, enum, unions
in function signatures
* add .h file generation for extern struct, enum, unions
These are on the slower side and could be improved. No performance optimizations
yet have been done.
```
Cpu: Intel(R) Core(TM) i5-6500 CPU @ 3.20GHz
```
-- Sha3-256
```
Zig --release-fast
93 Mb/s
Zig --release-safe
99 Mb/s
Zig
4 Mb/s
```
-- Sha3-512
```
Zig --release-fast
49 Mb/s
Zig --release-safe
54 Mb/s
Zig
2 Mb/s
```
Interestingly, release-safe is producing slightly better code than
release-fast.
* error return tracing is disabled in release-fast mode
* add @errorReturnTrace
* zig build API changes build return type from `void` to `%void`
* allow `void`, `noreturn`, and `u8` from main. closes#535
Some performance comparisons to C.
We take the fastest time measurement taken across multiple runs.
The block hashing functions use the same md5/sha1 methods.
```
Cpu: Intel(R) Core(TM) i5-6500 CPU @ 3.20GHz
Gcc: 7.2.1 20171224
Clang: 5.0.1
Zig: 0.1.1.304f6f1d
```
See https://www.nayuki.io/page/fast-md5-hash-implementation-in-x86-assembly:
```
gcc -O2
661 Mb/s
clang -O2
490 Mb/s
zig --release-fast and zig --release-safe
570 Mb/s
zig
50 Mb/s
```
See https://www.nayuki.io/page/fast-sha1-hash-implementation-in-x86-assembly:
```
gcc -O2
588 Mb/s
clang -O2
563 Mb/s
zig --release-fast and zig --release-safe
610 Mb/s
zig
21 Mb/s
```
In short, zig provides pretty useful tools for writing this sort of
code. We are in the lead against clang (which uses the same LLVM
backend) with us being slower only against md5 with GCC.
* better error message for realpath failing
* fix bug in std.io.readFileAllocExtra incorrectly returning
error.EndOfStream
* implement std.os.selfExePath and std.os.selfExeDirPath for windows
Before:
* IR basic blocks are in arbitrary order
* when doing an IR pass, when a block is encountered, code
must look at all the instructions in the old basic block,
determine what blocks are referenced, and queue up those
old basic blocks first.
* This had a bug (See #667)
Now:
* IR basic blocks are required to be in an order that guarantees
they will be referenced by a branch, before any instructions
within are referenced.
ir pass1 is updated to meet this constraint.
* When doing an IR pass, we iterate over old basic blocks
in the order they appear. Blocks which have not been
referenced are discarded.
* After the pass is complete, we must iterate again to look
for old basic blocks which now point to incomplete new
basic blocks, due to comptime code generation.
* This last part can probably be optimized - most of the time
we don't need to iterate over the basic block again.
closes#667
...field to const slice parameter
we use a packed struct internally to represent a const array
of disparate union values, and needed to update the internal
getelementptr instruction to recognize that.
closes#664
closes#346closes#630
regression: translate-c can no longer translate switch statements.
after #629 we can ressurect and modify the code to utilize arbitrarily
returning from blocks.
introduces the `@export` builtin function which can be used
in a comptime block to conditionally export a function.
it also allows creation of aliases.
previous export syntax is still allowed.
closes#462closes#420
* add @noInlineCall - see #640
This fixes a crash in --release-safe and --release-fast modes
where the optimizer inlines everything into _start and
clobbers the command line argument data.
If we were able to verify that the user's code never reads
command line args, we could leave off this "no inline"
attribute.
* add i29 and u29 primitive types. u29 is the type of alignment,
so it makes sense to be a primitive.
probably in the future we'll make any `i` or `u` followed by
digits into a primitive.
* add `aligned` functions to Allocator interface
* add `os.argsAlloc` and `os.argsFree` so that you can get
a `[]const []u8`, do whatever arg parsing you want, and then free
it. For now this uses the other API under the hood, but it could
be reimplemented to do a single allocation.
* add tests to make sure command line argument parsing works.
* @enumTagName renamed to @tagName and it works on enums and
union-enums
* Remove the EnumTag type. Now there is only enum and union,
and the tag type of a union is always an enum.
* unions support specifying the tag enum type, and they support
inferring an enum tag type.
* Enums no longer support field types but they do support
setting the tag values. Likewise union-enums when inferring
an enum tag type support setting the tag values.
* It is now an error for enums and unions to have 0 fields.
* switch statements support union-enums
closes#618
* rename decode to decodeExactUnsafe.
* add decodeExact, which checks for invalid chars and padding.
* add decodeWithIgnore, which also allows ignoring chars.
* alphabets are supplied to the decoders with their
char-to-index mapping already built, which enables it to be
done at comptime.
* all decode/encode apis except decodeWithIgnore require dest
to be the exactly correct length. This is calculated by a
calc function corresponding to each api. These apis no longer
return the dest parameter.
* for decodeWithIgnore, an exact size cannot be known a priori.
Instead, a calc function gives an upperbound, and a runtime
error is returned in case of overflow. decodeWithIgnore
returns the number of bytes written to dest.
closes#611
* fix fstat wrong on darwin
* move std.debug.global_allocator to std.debug.global_allocator_state and make it private
* add std.debug.global_allocator as a pointer (to upgrade your zig code remove
the '&')
I started working on #465 and made some corresponding std.io
API changes.
New structs:
* std.io.FileInStream
* std.io.FileOutStream
* std.io.BufferedOutStream
* std.io.BufferedInStream
Removed:
* std.io.File.in_stream
* std.io.File.out_stream
Now instead of &file.out_stream or &file.in_stream to get access to
the stream API for a file, you get it like this:
var file_in_stream = io.FileInStream.init(&file);
const in_stream = &file_in_stream.stream;
var file_out_stream = io.FileOutStream.init(&file);
const out_stream = &file_out_stream.stream;
This is evidence that we might not need any OOP features -
See #130.
Andrew Kelley
Marc Tiehuis
Ben Noordhuis
Jeff Fowler
Andrea Orru
Jimmi Holst Christensen
Josh Wolfe
Andrea Orru
Jimmi Holst Christensen
Peter Rönnquist
dimenus
Brendon Scheinman
scurest