Differences in regex support between gcc 4.9.2 and gcc 5.3

Question

Can anyone more familiar with gcc point out why the sample below fails to match on gcc 4.9.2 but succeeds on gcc 5.3? Is there anything I can do to alternate the pattern so that it would work (also seems to work fine on VS 2013)?

#include <iostream>
#include <regex>

std::regex pattern("HTTP/(\\d\\.\\d)\\s(\\d{3})\\s(.*)\\r\\n(([!#\\$%&\\*\\+\\-\\./a-zA-Z\\^_`\\|-]+\\:[^\\r]+\\r\\n)*)\\r\\n");

const char* test = "HTTP/1.1 200 OK\r\nHost: 192.168.1.72:8080\r\nContent-Length: 86\r\n\r\n";

int main()
{
    std::cmatch results;
    bool matched = std::regex_search(test, test + strlen(test), results, pattern);
    std::cout << matched;
    return 0; 
}

I assume I am using something that is not supported in gcc 4.9.2 but was added on or fixed later, but I have no idea where to look it up.

UPDATE

Due to the amount of help and suggestions I tried to backtrack the issue instead of just switching to gcc 5. I get correct matches with this modification:

#include <iostream>
#include <regex>

std::regex pattern("HTTP/(\\d\\.\\d)\\s(\\d{3})\\s(.*?)\\r\\n(?:([^:]+\\:[^\\r]+\\r\\n)*)\\r\\n");

const char* test = "HTTP/1.1 200 OK\r\nHost: 192.168.1.72:8080\r\nContent-Length: 86\r\n\r\n";

int main()
{
    std::cmatch results;
    bool matched = std::regex_search(test, test + strlen(test), results, pattern);
    std::cout << matched << std::endl;
    if (matched)
    {
        for (const auto& result : results)
        {
            std::cout << "matched: " << result.str() << std::endl;
        }
    }
    return 0;
}

So I guess the problem is with the group that matches the HTTP header name. Will check further.

UPDATE 2

std::regex pattern(R"(HTTP/(\d\.\d)\s(\d{3})\s(.*?)\r\n(?:([!#$&a-zA-Z^_`|-]+\:[^\r]+\r\n)*)\r\n)")

is the last thing that works. Adding any of the remaining characters that I had in my group - %*+-. (escaped or not epscaped) - breaks it.

possibly related to: http://stackoverflow.com/questions/15059162/c11-regex-matching — NathanOliver, Apr 19 '16 at 13:44
As a sidenote, I recommend raw string literals to limit the onslaught of backslashes. — eerorika, Apr 19 '16 at 13:52
@NathanOliver I did not quite catch that - it means the version of `libstdc++` I use with gcc 4.9.2 is to be blamed? — Rudolfs Bundulis, Apr 19 '16 at 14:09
@RudolfsBundulis It is possible. The answer says it should be fixed by gcc 4.9 but there could be bugs in 4.9.2 since the implementation was still new. — NathanOliver, Apr 19 '16 at 14:10
@user2079303 thanks, hadn't managed to adapt them yet, but it clearly makes things prettier:) — Rudolfs Bundulis, Apr 19 '16 at 14:19
@Laurel - thanks for the help, to save the time and answer all the suggestions I posted the last modification that still matches. Seems that the HTTP header name matching part is the pitfall. Not sure why. — Rudolfs Bundulis, Apr 19 '16 at 16:28
@RudolfsBundulis, which part(s) are trying to match/extract with the regex? — Super-intelligent Shade, Apr 21 '16 at 01:58
All the HTTP response metadata - HTTP version, method, path and headers. — Rudolfs Bundulis, Apr 21 '16 at 08:13
[You may have been bitten on the butt by a known deficiency in G++](https://gcc.gnu.org/onlinedocs/libstdc++/manual/status.html#status.iso.2011). I'm not sure, though, since I don't really know how the regex library works, just that it does in G++5.x. In the mean time, if you can't upgrade, use Boost.Regex; it supports more stuff, anyway. — Nic, Jun 23 '16 at 03:31

score 2 · Answer 1 · answered Oct 04 '18 at 23:43

So I know GCC did not support the c++11 regex library until GCC 4.9 officially. See Is gcc 4.8 or earlier buggy about regular expressions?. Since it was so new, it is likely that it had a few bugs to smooth out. Pinning down the exact cause would be difficult, but the problem is in the implementation and not in the regex.

Side note: I remember spending 20 minutes one time trying to figure out what was wrong with my regex when I found the mentioned article and realized that I was using gcc 4.8.*. Since the machine I had to run on wasn't mine, I basically ended up compiling on a different, similar platform with a later version of gcc and a few hacks and then it ran on the target platform.

Differences in regex support between gcc 4.9.2 and gcc 5.3

1 Answers1