Le 20/11/2015 19:43, Georg Baum a écrit :
Scott Kostyshak wrote:
I was hoping the recent commit fixing regex issues would fix the failing
test (test_biblio) of 'make check'. I just wanted to confirm that the
test fails for others on current master also.
It does. The reason is that the output of escape_special_chars() changes
when compiling with std::regex. The attached patch would fix the test for
std::regex, but then make check would fail when using boost::regex.
Of course, the source should be corrected, not the test.
BTW I read the new documentation for tests and it did not help me at
all. It said nothing for "make check", and when I run "make check" it is
still a mystery to me where I can get any log file that says what went
wrong. (The file test_biblio.log is always empty.)
This shows that the regex engines we use do still interpret the expressions
in different ways, and I believe I also know the reason: The expressions
have changed to ECMAScript syntax, but the boost::regex engine does not
default to ECMAScript, but to boost perl style.
But "perl" for Boost is just a synonym for "ECMAScript", and the
difference between std and boost in the current master is only
supposed to be a few perl extensions added. (See
<http://www.boost.org/doc/libs/1_57_0/libs/regex/doc/html/boost_regex/ref/syntax_option_type/syntax_option_type_perl.html>.)
Therefore the problem is probably still in the expressions. What I
understand from your patch is that the output in std::regex is still
wrong: the regex adds \\ instead of \. The result of the test you showed
with the patch. I am worried by the line:
return lyx::regex_replace(expr, reg, "\\\\$&");
This should be:
return lyx::regex_replace(expr, reg, "\\$&");
but then this would no longer work with Boost, because it interprets the
backslash as an escape sequence... This appears to be a bug in Boost:
format_default: Specifies that when a regular expression match is to
be replaced by a new string, that the new string is constructed using
the rules used by the ECMAScript replace function in ECMA-262,
ECMAScript Language Specification, Chapter 15 part 5.4.11
String.prototype.replace. (FWD.1).
This is functionally identical to the Perl format string rules.
But the "Perl format" treats backslashes as a character one must escape,
which contradicts the preceding sentence.
Therefore, we do now have
exactly the inverted situation than before dbce5cafcc167: regexes are
interpreted correctly when using std::regx, but not when using boost::regex.
On my side master still works fine with Boost so I don't see an inverted
situation. But, if we did the correct thing and changed the format
string to \$& as it should be, we would indeed end up in a symmetric
situation.
What needs to be done now is to call boost::regex in ECMAScript mode. Then
escape_special_chars() will produce the same output for both regex engines.
As I understand, both are already set to ECMAScript, the problem being
that Boost does not implement ECMAScript.
What I am not 100% sure yet is whether it is correct, there seem to be too
many backslashes.
I am afraid that for now there are only two unpleasant solutions:
* fix this particular replacement using a hack, or
* use Boost on every platform until we can get rid of it entirely.
Guillaume
