operator<< for boost::posix_time::ptime ignores ostream flags

[adam@…]

Consider the following:

boost::posix_time::ptime a;

cout << "one " << 1 << endl;
cout.clear(ios::failbit);
cout << "two " << 2 << endl;
cout << "time " << a << endl;
cout << "three" << 3 << endl;
cout.clear(ios::goodbit);
cout << "four " << 4 << endl;

Since cout should not produce output when ios::failbit is set, one would expect the output to be:

one 1
four 4

However, the output is:

one 1
not-a-date-timefour 4

[christian.stimming@…]

I second this bug. The error still exists in boost-1.43.0. I'll attach a test case and a patch that fixes the bug by querying ostream::fail() before doing anything in the operator<<.

[christian.stimming@…]

Test case for this bug: Due to the error, in the second line the date still appears even though it shouldn't have been printed because of the ios::failbit.

[christian.stimming@…]

Patch for fixing this bug by querying ostream::fail() before doing anything in the operator<< of ptime and time_duration

[christian.stimming@…]

Better patch for this error which (after reading my Stroustrup again) is using the ostream's sentry constructor to check for potential failbits.

[Katie Chan]

The most recent patch when applied to 1.45 does not produce the "expected output".

[christian.stimming@…]

I don't understand ktchan's comment here. When compiling the attached example ptime-operator.cpp with any current boost (1.43, 1.45, SVN trunk; it doesn't matter which one), the output is

Should appear: Text. And Date: 2010-Dec-21 14:12:24.398138 Boost v104300
2010-Dec-21 14:12:24.398138
This line should appear (but no single date above): Text. And Date: 2010-Dec-21 14:12:24.398138

which is erroneous - the second line should have been empty. After applying the attached second patch fix-ptime-operator-2nd-bug1574.patch, the output is

Should appear: Text. And Date: 2010-Dec-21 14:12:32.328110 Boost v104300

This line should appear (but no single date above): Text. And Date: 2010-Dec-21 14:12:32.328110

which is the correct and expected output. Can you please apply the patch?

[Katie Chan]

Hmm, my bad. I swear that when I tested it last time round, the output wasn't what was expected. :?

[Rob Stewart]

The patch looks good except for the ill-formed identifiers with double underscores. I also find "cerb" to be nonsensical. The following changes would be better:

Instead of:

typedef std::basic_ostream<CharT, TraitsT> __ostream_type;
typename __ostream_type::sentry cerb(os);
if (cerb)

Use this:

typename std::basic_ostream<CharT, TraitsT>::sentry ready(os);
if (ready)

[christian.stimming@…]

Feel free to commit the patch in the changed form you explained. However, the form I submitted was copied verbatim from gcc's STL, on Linux available in the file /usr/include/c++/4.4/bits/ostream_insert.h line 77ff.

I don't know specific reasons for choosing the form I copied from there, so feel free to implement it differently in boost. But if you ask me, I'd stick to the form proposed by gcc's STL.

[Steven Watanabe]

The reason the STL uses is exactly the reason that we can't use it. Identifiers containing a double underscore are reserved to the implementation (i.e. the compiler and standard library) for any use.

[Dean Michael Berris]

After looking at the ticket comments above I've made some modifications to make it a little more palatable playing nicely with the standard's forbidding the use of double underscores in identifiers that are not defined by the implementation. Please see the attached new modified patch.

[Dean Michael Berris]

Applies cleanly to trunk 68583

[christian.stimming@…]

Error still present in boost-1.46 and also still SVN trunk. Any news on applying the patch?

[christian.stimming@…]

Error still present in boost-1.52 and also still SVN trunk. Any news on applying the patch?

[christian.stimming@…]

Error still present in boost-1.53 and also still SVN trunk. Any news on applying the patch?