disable fuzzer on 32 bit linux #23008
Conversation
This follows a commit that disallows fuzzing on 32-bit platforms, but it still has to build given the current build logic.
lib/compiler/build_runner.zig
Outdated
| // More specifically, posix.mmap()'s second parameter, `length: usize`, | ||
| // must be the same bit-size as std.fs.getEndPos()'s return value, which is `u64`. | ||
| // Affects or affected by issues #5185, #22523, and #22464. | ||
| 64 => {}, |
There was a problem hiding this comment.
This allows fuzz testing only on 64-bit platforms, but it's not enough to fix the build runner on 32-bit platforms.
lib/std/Build/Fuzz/WebServer.zig
Outdated
| log.err("coverage file '{}' is too large to mmap", .{coverage_file_path}); | ||
| return error.OutOfMemory; | ||
| } | ||
| const file_usize: usize = @truncate(file_size); |
There was a problem hiding this comment.
This @truncate is NOP for 64-bit platforms but necessary for the build to work on 32-bit platforms, even though we don't allow running fuzz tests of 32-bit platforms.
There was a problem hiding this comment.
But why? This code shouldn't be compiled for 32-bit.
There was a problem hiding this comment.
I agree, but where is the build logic to exclude it on 32-bit?
There was a problem hiding this comment.
The lines that I pointed out earlier are a chokepoint, so the patch there will make all of WebServer.zig dead code.
There was a problem hiding this comment.
My mistake--a made a goof while testing.
I've removed these changes. Would you like a new PR omitting the changeset introducing changes to WebServer.zig?
lib/std/Build/Fuzz/WebServer.zig
Outdated
| @@ -628,9 +628,17 @@ fn prepareTables( | |||
| return error.AlreadyReported; | |||
| }; | |||
|
|
|||
| // Account for bit-size mismatch between fs.getEndPos() and posix.mmap() ideas about file length | |||
| if (file_size >= std.math.maxInt(usize)) { | |||
There was a problem hiding this comment.
This test is not necessary so long as we prohibit fuzz testing on 32-bit platforms.
However, it shows one possible way of safely trying it out on those platforms. The actual test could be affected, for libc targets, by -D_FILE_OFFSET_BITS. Raspbian 32-bit man 2 stat notes:
EOVERFLOW
pathname or fd refers to a file whose size, inode number, or number of blocks cannot be represented
in, respectively, the types off_t, ino_t, or blkcnt_t. This error can occur when, for example, an
application compiled on a 32-bit platform without -D_FILE_OFFSET_BITS=64 calls stat() on a file
whose size exceeds (1<<31)-1 bytes.
Addresses #22523 by: