[llvm-bugs] [Bug 36099] New: Input stream formatted input for `float` produces incorrect result for non-hexadecimal-prefixed input containing hexadecimal characters without an exponent

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=36099

            Bug ID: 36099
           Summary: Input stream formatted input for `float` produces
                    incorrect result for non-hexadecimal-prefixed input
                    containing hexadecimal characters without an exponent
           Product: libc++
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: All Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: brycelelbach at gmail.com
                CC: llvm-bugs at lists.llvm.org, mclow.lists at gmail.com

=========
Test Case
=========

    #include <sstream>
    #include <iostream>

    int main() {
        std::istringstream s("1.000f32");

        float f;
        s >> f;

        std::cout << f << std::endl;
    }

==================
Steps to Reproduce
==================

https://wandbox.org/permlink/3d9Z9Fr6vqzcx29w

===============
Observed Output
===============

0

===============
Expected Output
===============

1

This appears to be a bug in libc++'s hexadecimal `float` formatted input.

libc++'s formatted stream input implementation for `float` calls `num_get`
(ISO/IEC 14882 [istream.formatted.arithmetic]).

In the case of `float`, `num_get` accumulates valid characters until it
encounters an invalid character (stage 1), and then does the equivalent of
calling the C library function `strtof` on the accumulated string (ISO/IEC
14882 [facet.num.get.virtuals] (3.3.3) "Stage 2"). My guess is that in this
case,
libc++ considers the characters `[a-z]` (hexadecimal digits) to be valid in
addition to `[0-9+-eE]` in stage 1, but in stage 2 (the equivalent-to-`strtof`
stage), the string "1.000f32" is rejected because it doesn't have the
hexadecimal prefix (e.g. it's not "0x1.000f32").

Feeding the entire string directly into `strtof` consumes "1.000" and returns
1:

https://wandbox.org/permlink/BRWaxfmoTNewQMTo

If you change the input string to contain a character that is not a hexadecimal
digit (e.g. "1.000g32"), libc++ returns 1:

https://wandbox.org/permlink/BRWaxfmoTNewQMTo

libc++ should only accept hexadecimal digits in floats when the hexadecimal
prefix is present. Otherwise, there are ambiguities. Consider "1.000e32" - is
it a hexadecimal `float`, or a `float` with an exponent.

libstdc++ and MSVC print 1 for the test case:

libstdc++: https://wandbox.org/permlink/4EF17CZxau2CrYTA
MSVC: http://rextester.com/STNGQX93481

===========
Environment
===========

Platform: Godbolt (Linux Docker)
Clang/LLVM Version: Top of trunk as of 01/25/2018

-- 
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/20180125/47297be2/attachment.html>