There's an unfortunate footgun in the current design of error sets.
The debug info type for every error set is the same as the debug info
type of the global error set, which is essentially an enum forward
declaration. The problem is that when we "replace" the forward
declaration with the final value, once we know all the possible errors,
we have to update the pointers of every error set.
So the footgun is that if you ever copy the debug info type of the
global error set, you have to add the address of the pointer to a list
of pointers that need to be updated once we "replace" the forward
declaration. I activated the footgun when I introduced the optimization
that `?anyerror` types are the same size as `anyerror` types (using 0 as
the null value), because I introduced a pointer copy of the global error
set debug info type, but forgot to add it to the list.
I'm sure that there is a better way to code this, which does not have
the footgun, but this commit contains only a fix, not a reworking of the
logic.
closes#1937
this would work if @llvm.sadd.with.overflow supported
vectors, which it does in trunk. but it does not support
them in llvm 7 or even in llvm 8 release branch.
so the next commit after this will have to do a different
strategy, but when llvm 9 comes out it may be worth coming
back to this one.
This commit contains everything from the copy-elision-2
branch that does not have to do with copy elision directly,
but is generally useful for master branch.
* All const values know their parents, when applicable, not
just structs and unions.
* Null pointers in const values are represented explicitly,
rather than as a HardCodedAddr value of 0.
* Rename "maybe" to "optional" in various code locations.
* Separate DeclVarSrc and DeclVarGen
* Separate PtrCastSrc and PtrCastGen
* Separate CmpxchgSrc and CmpxchgGen
* Represent optional error set as an integer, using the 0 value.
In a const value, it uses nullptr.
* Introduce type_has_one_possible_value and use it where applicable.
* Fix debug builds not setting memory to 0xaa when storing
undefined.
* Separate the type of a variable from the const value of a variable.
* Use copy_const_val where appropriate.
* Rearrange structs to pack data more efficiently.
* Move test/cases/* to test/behavior/*
* Use `std.debug.assertOrPanic` in behavior tests instead of
`std.debug.assert`.
* Fix outdated slice syntax in docs.
* bitreverse - give bswap behavior
* bitreverse, comptime_ints, negative values still not working?
* bitreverse working for negative comptime ints
* Finished bitreverse test cases
* Undo exporting a bigint function. @bitreverse test name includes ampersand
* added docs entry for @bitreverse
* add `@bswap` builtin function. See #767
* comptime evaluation facilities are improved to be able to
handle a `@ptrCast` with a backing array.
* `@truncate` allows "truncating" a u0 value to any integer
type, and the result is always comptime known to be `0`.
* when specifying pointer alignment in a type expression,
the alignment value of pointers which do not have addresses
at runtime is ignored, and always has the default/ABI alignment
* threw in a fix to freebsd/x86_64.zig to update syntax from
language changes
* some improvements are pending #863closes#638closes#1733
std lib API changes
* io.InStream().readIntNe renamed to readIntNative
* io.InStream().readIntLe renamed to readIntLittle
* io.InStream().readIntBe renamed to readIntBig
* introduced io.InStream().readIntForeign
* io.InStream().readInt has parameter order changed
* io.InStream().readVarInt has parameter order changed
* io.InStream().writeIntNe renamed to writeIntNative
* introduced io.InStream().writeIntForeign
* io.InStream().writeIntLe renamed to writeIntLittle
* io.InStream().writeIntBe renamed to writeIntBig
* io.InStream().writeInt has parameter order changed
* mem.readInt has different parameters and semantics
* introduced mem.readIntNative
* introduced mem.readIntForeign
* mem.readIntBE renamed to mem.readIntBig and different API
* mem.readIntLE renamed to mem.readIntLittle and different API
* introduced mem.readIntSliceNative
* introduced mem.readIntSliceForeign
* introduced mem.readIntSliceLittle
* introduced mem.readIntSliceBig
* introduced mem.readIntSlice
* mem.writeInt has different parameters and semantics
* introduced mem.writeIntNative
* introduced mem.writeIntForeign
* mem.writeIntBE renamed to mem.readIntBig and different semantics
* mem.writeIntLE renamed to mem.readIntLittle and different semantics
* introduced mem.writeIntSliceForeign
* introduced mem.writeIntSliceNative
* introduced mem.writeIntSliceBig
* introduced mem.writeIntSliceLittle
* introduced mem.writeIntSlice
* removed mem.endianSwapIfLe
* removed mem.endianSwapIfBe
* removed mem.endianSwapIf
* added mem.littleToNative
* added mem.bigToNative
* added mem.toNative
* added mem.nativeTo
* added mem.nativeToLittle
* added mem.nativeToBig
* add __multi3 compiler rt function. See #1290
* compiler rt includes ARM functions for thumb and aarch64 and
other sub-arches left out. See #1526
* support C ABI for returning structs on ARM. see #1481
We were caching the ConstExprValue of string literals,
which works if you can never modify ConstExprValues.
This premise is broken with `comptime var ...`.
So I implemented an optimization in ConstExprValue
arrays, where it stores a `Buf *` directly rather
than an array of ConstExprValues for the elements,
and then similar to array of undefined, it is
expanded into the canonical form when necessary.
However many operations can happen directly on the
`Buf *`, which is faster.
Furthermore, before a ConstExprValue array is expanded
into canonical form, it removes itself from the string
literal cache. This fixes the issue, because before an
array element is modified it would have to be expanded.
closes#1076
