The WebAssembly spec requires signed LEB128 to be encoded up to a maximum number of bytes (max 5 bytes for i32, max 10 bytes for i64) and that "unused" bits are all 0 if the number is positive and all 1 if the number is negative. The Zig LEB128 implementation already enforces the max number of bytes and does check the unused bytes https://github.com/ziglang/zig/blob/master/lib/std/leb128.zig#L70-L79.
However, the WebAssembly test suite has a number of tests that were failing validation (expecting the wasm module to fail validation, but when running the tests, those examples were actually passing validation):
https://github.com/malcolmstill/foxwren/blob/master/test/testsuite/binary-leb128.wast#L893-L902https://github.com/malcolmstill/foxwren/blob/master/test/testsuite/binary-leb128.wast#L934-L943
Notably the failures are both cases of negative numbers and the top 4 bits of the last byte are zero. And I believe this is the issue: we're only currently checking the "unused" / remaining_bits if we overflow, but in the case of 0x0_ no overflow happens and so the bits go unchecked.
In other words:
\xff\xff\xff\xff\7f rightly successfully decodes (because it overflows and the remaining bits are 0b1111)
\xff\xff\xff\xff\6f rightly errors with overflow (because it overflows and the remaining bits are 0b1110)
\xff\xff\xff\xff\0f incorrectly decodes when it should error (because the top 4 bits are all 0, and so no overflow occurs and no check that the unused bits are 1 happens)
This PR adds a the remaining_bits check in an else branch of the @shlWithOverflow when we're looking at the last byte and the number being decoded is negative.
Note: this means a couple of the test cases in leb128.zig that are down as decoding shouldn't actually decode so I added the appropriate 1 bits.
I believe these are Linux specific so they will need to be os-gated
in `elf.zig` at some point, but I reckon it should be fine to have
them as-is right now since the ELF linker work will mainly be done
on x86-64 Linux at first.
* test runner is improved to respect `error.SkipZigTest`
* start code is improved to `@setAlignStack(16)` before calling main()
* the newly passing behavior test has a workaround for the fact that
stage2 cannot yet call `std.Target.x86.featureSetHas()` at comptime.
This is blocking on comptime closures. The workaround is that there
is a new decl `@import("builtin").stage2_x86_cx16` which is a `bool`.
* Implement `@setAlignStack`. This language feature should be re-evaluated
at some point - I'll file an issue for it.
* LLVM backend: apply/remove the cold attribute and noinline attribute
where appropriate.
* LLVM backend: loads and stores are properly annotated with alignment
and volatile attributes.
* LLVM backend: allocas are properly annotated with alignment.
* Type: fix integers reporting wrong alignment for 256-bit integers and
beyond. Once you get to 16 byte aligned, there is no further
alignment for larger integers.
This commit reverts 6d37ae95edc06f15e4e77f64e8e637dd5d269183 and
8f8294a809f9d975735377e7bfcc2c47ccfc4cb7. I don't know why they caused a
failure but that investigation can happen while the CI is green.
This supports the case when it is known that LLVM, Clang, LLD were built
with Clang (or `zig c++`). This commit updates the Linux CI script to
pass this since we build using a zig tarball.
Investigating hexops/zorex#4, I found that `--test-evented-io` is currently broken in
the latest Zig nightly. See #9779 for a small reproduction.
The issue is that allocation errors here are not correctly handled, as this function
returns `void` and all other error cases `@panic`, the allocation failure should also
use `@panic`.
Fixes#9779
Helps hexops/zorex#4
Signed-off-by: Stephen Gutekanst <stephen@hexops.com>
* In watch mode, when changing the C source, we will trigger complete
relinking of objects, dylibs and archives (atoms coming from the
incremental updates stay put however). This means, we need to undo
metadata populated when linking in objects, archives and dylibs.
* Remove unused splitting section into atoms bit. This optimisation
will probably be best rewritten from scratch once self-hosted
matures so parking the idea for now. Also, for easier management
of atoms spawned from the Object file, keep the atoms subgraph as
part of the Object file struct.
* Remove obsolete ref to static initializers in object struct.
* Implement handling of global symbol collision in updateDeclExports.
Some systems (Solaris, OpenBSD, AIX) change their definitions of
sockaddr_storage to be larger than 128 bytes. This comment adds a new
constant in the `sockaddr` that defines the size for every system.
Fixes#9759
The entire 'path' array would get written to the formatting function,
when it should instead be treated as a regular zero-terminated string.
Note that this doesn't handle abstract paths on Linux, those paths
*start* with a \0 byte and are hence treated as empty strings instead.
But fixing that would require more adjustments than just formatting, in
particular to getOsSockLen().
* langref: add some more "see also" links for atomics
* Add the following AIR instructions
- atomic_load
- atomic_store_unordered
- atomic_store_monotonic
- atomic_store_release
- atomic_store_seq_cst
- atomic_rmw
* Implement those AIR instructions in LLVM and C backends.
* AstGen: make the `ty` result locations for `@atomicRmw`, `@atomicLoad`,
and `@atomicStore` be `coerced_ty` to avoid unnecessary ZIR
instructions when Sema will be doing the coercions redundantly.
* Sema for `@atomicLoad` and `@atomicRmw` is done, however Sema for
`@atomicStore` is not yet implemented.
- comptime eval for `@atomicRmw` is not yet implemented.
* Sema: flesh out `coerceInMemoryAllowed` a little bit more. It can now
handle pointers.
The following is an azure failure that occured Sep 13:
del : Cannot remove item D:\a\1\s\sfx.exe: The process cannot access the file 'D:\a\1\s\sfx.exe' because it is being used by another process.
Windows will keep a hold of recently run exeutables even after their process has exited. To avoid this I've just removed the deletion of the exe file. It's about 70 MB so it's probably OK.
* Added fseeki64.c from mingw-w64 9.0.0. This file was missing in Zig distribution. This file contains implementation for _fseeki64 and _ftelli64 functions.
* Implement Sema for `@cmpxchgWeak` and `@cmpxchgStrong`. Both runtime
and comptime codepaths are implement.
* Implement Codegen for LLVM backend and C backend.
* Add LazySrcLoc.node_offset_builtin_call_argX 3...5
* Sema: rework comptime control flow.
- `error.ComptimeReturn` is used to signal that a comptime function
call has returned a result (stored in the Inlining struct).
`analyzeCall` notices this and handles the result.
- The ZIR instructions `break_inline`, `block_inline`,
`condbr_inline` are now redundant and can be deleted. `break`,
`block`, and `condbr` function equivalently inside a comptime scope.
- The ZIR instructions `loop` and `repeat` also are modified to
directly perform comptime control flow inside a comptime scope,
skipping an unnecessary mechanism for analysis of runtime code.
This makes Zig perform closer to an interpreter when evaluating
comptime code.
* Sema: zirRetErrValue looks at Sema.ret_fn_ty rather than sema.func
for adding to the inferred error set. This fixes a bug for
inlined/comptime function calls.
* Implement ZIR printing for cmpxchg.
* stage1: make cmpxchg respect --single-threaded
- Our LLVM C++ API wrapper failed to expose this boolean flag before.
* Fix AIR printing for struct fields showing incorrect liveness data.
* add functions to decode an epoch timestamp
The code added here is alternative to the libc gmtime function. This function takes a unix epoch timestamp and decodes it into things like the year/day/time/etc. I looked at various libc implementations to see how it was implemented and this appears to be correct. I reorganized it so that applications can choose which data they need rather than calcualting it all in a single function. The data structures layout the order of operations required to decode various things like the year/month or time of day.
* set Month.jan to 1 instead of 0 to avoid +1 in conversion to numeric
* add another test
* remove unnecessary comptimeMod
This applies to stage2 where we make use of the cache system to work
out if we need to relink objects when performing incremental updates.
When the process is restarted however, while in principle the idea is
to carry on where we left off by reparsing the prelinked binary from
file, the required machinery is not there yet, and therefore we always
fully relink upon restart.
