Non-Arm CPUs use u16 as the parameter to __extendhfxf2 and the return value of
__truncxfhf2, so insert appropriate bitcasts in gen_soft_f80_widen_or_shorten.
Otherwise, LLVM might crash because the functions are called in a different way
than its compiler-rt definition.
This fixes stage1 build on SPARCv9, and possibly other non-x86, non-Arm CPUs.
Instead of assuming that the zig version used with `zig build` is
appropriate for building the self-hosted compiler component, we follow
the same path as the cmake build script and build zig0, using that to
produce zig1.o, re-linking to produce stage1.
This allows an arbitrarily old Zig version and corresponding build.zig
script to build all future versions of Zig from source via the bootstrap
compiler. In other words, it allows us to use `zig build` as an
alternative to cmake when bootstrapping.
std.fmt.parseHexFloat allow both 0x and 0X as prefix, so in order to
keep things consistent. This commit modifies std.fmt.parseWithSign to
check for the prefix case insensitively in order to allow both upper
case and lower case prefix.
This change now allows: 0O, 0B and 0X as prefixes for parsing.
* move DWARF in file if LINKEDIT spilled in dSYM
* update VM addresses for all segments
* introduce copyFileRangeOverlappedAll instead of File.copyRangeAll
since we make lots of overlapping writes in MachO linker
This "feature" of gcc/clang means to treat this as a positional
link object, but using the library search directories as a prefix.
We solve this problem in the CLI layer, using a separate map for
the data since it is an uncommon case.
Closes#10851
On Windows, `argv` is not populated by start code, and instead left as undefined. This is problematic, and can lead to incorrect programs compiling, but panicking when trying to access `argv`. This change causes these programs to produce a compile error on Windows instead, which is far preferable to a runtime panic.
* update number of type abbrevs to match Elf linker
* update `DebugSymbols` to write symbol and string tables
at the end to match the `MachO` linker
* TODO: update segment vm addresses when growing segments in
the binary
* TODO: store DWARF relocations in linker's interned arena
Previously, `suffix` was copied to `output_buffer` at position
`max_end`, thereby writing into reserved space after `max_end`.
This only worked because `suffix` was not larger than
`bytes_needed_for_esc_codes_at_end` (otherwise there'd be a potential
buffer overrun) and no escape codes at end are actually written.
Since 2d5b2bf1c986d037ef965bf8c9b4d8dfd5967478, escape codes are no
longer written to the end of the buffer. They are now written
exclusively to the front of the buffer.
This allows removing `bytes_needed_for_esc_codes_at_end` and
simplifying the suffix printing logic.
This also fixes the bug that the ellipse suffix was not printed in
Windows terminals because `end.* > max_end` was never true.
In #10859 I moved the `test_node.end()` call after everything else has
been logged. Now the `test_fn.name` is printed by `Progress` itself,
making the additional log obsolete.
The information whether a register is allocated to an instruction is
already encoded in the free_registers "bitmap". Duplicating that
information in the registers map is unnecessary and may lead to
performance degradations.
