[LLVMbugs] [Bug 21506] New: libc++ should have its own classic_table
bugzilla-daemon at llvm.org
bugzilla-daemon at llvm.org
Thu Nov 6 13:24:37 PST 2014
http://llvm.org/bugs/show_bug.cgi?id=21506
Bug ID: 21506
Summary: libc++ should have its own classic_table
Product: libc++
Version: unspecified
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: All Bugs
Assignee: danalbert at google.com
Reporter: danalbert at google.com
CC: eric at efcs.ca, jonathan at codesourcery.com,
llvmbugs at cs.uiuc.edu, mclow.lists at gmail.com
Classification: Unclassified
We should probably be shipping our own classic_table instead of relying on one
from the underlying C library.
1. The C library isn't required to even provide such a table. IIRC, musl does
not.
2. The underlying type of the table is undefined, as are the values that map to
each mask. This is responsible for the #ifdef hell stew in [1].
3. Some C libraries (bionic, openbsd, and newlib, possibly others) use an 8-bit
mask for the table. It's still not 100% clear if this is okay, or if we need to
be able to define more than 8 values such that for any two values a and b, a &
b == 0. There are known failures in some libc++ tests [2] caused by this.
Defining our own table with a mask that is at least 16-bits wide (possibly 32)
would get us out of this mess.
I'll upload a patch soonish (once I find time) that does this unless anyone has
good reasons not to take this approach.
[1]: https://github.com/llvm-mirror/libcxx/blob/master/include/__locale#L324
[2]:
localization/locale.categories/category.ctype/locale.ctype.byname/is_1.pass.cpp
localization/locale.categories/category.ctype/locale.ctype.byname/scan_is.pass.cpp
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20141106/b4394ac8/attachment.html>
More information about the llvm-bugs
mailing list