Introduce 'allowSpecialFloats' for readers and 'useSpecialFloats' for writers, use consistent macro snprintf definition for writers and readers, provide new unit tests for #209
By not calling validate(), we can add
non-invasive features which will be simply ignored when user-code
is compiled against an old version. That way, we can often
avoid a minor version-bump.
The user can call validate() himself if he prefers that behavior.
This is an improper solution. If multiple Readers exist,
then the effect stackLimit is reduced because of side-effects.
But our options are limited. We need to address the security
hole without breaking binary-compatibility.
However, this is not likely to cause any practical problems because:
* Anyone using `operator>>(istream, Json::Value)` will be using the
new code already
* Multiple Readers are uncommon.
* The stackLimit is quite high.
* Deeply nested JSON probably would have hit the system limits anyway.
Tests are currently failing when git cloning on Windows with autocrlf = true. In
that setup multiline comments contain \r\n EOLs. The test code assumes that
comments contain \n EOLs and opens the .actual files (etc.) with "wt" which
converts \n to \r\n. Thus we end up with \r\r\n EOLs in the output, which
triggers a test failure.
Instead we should cannonicalize comments while reading so that they contain only
\n EOLs. This approach simplifies other parts of the reader and writer logic,
and requires no changes to the test. It is a breaking change, but probably the
Right Thing going forward.
This change also fixes dereferencing past the end of the comment string in
StyledWriter::writeCommentBeforeValue.
Tests should be added with appropriate .gitattributes for the input files to
ensure that we run tests for DOS, Mac, and Unix EOL files on all platforms. For
now this change is enough to unblock Windows builds.
issue #116
