problem with non-JIT pcre2

[0-wiz-0]

ucg compiled with pcre seems to work fine.
When compiled with pcre2-10.22, it builds fine but fails for me at runtime with:

path/to/ucg -r pcre path/to/ucg-source
/ucg: error: Error during regex parsing: PCRE2 JIT compilation error: bad JIT option

Not sure what this is about (and there's a weird '/' at the start of the line).
ucg --version output:

Compiler info:
 Name ($(CXX)): c++ -std=gnu++11
 Version string: "g++ (nb1 20160606) 5.4.0"

ISA extensions in use:
 sse4.2: yes
 popcnt: yes

libpcre info:
 Not linked against libpcre.

libpcre2-8 info:
 Version: 10.22 2016-07-29
 JIT support built in?: no
 JIT target architecture: none
 Newline style: LF

pcre2's self tests run fine.

[gvansickle]

Hi @0-wiz-0 ,
Does this happen with any regex/literal, or just with 'pcre'. I.e., does it work at all for you with PCRE2?

[0-wiz-0]

This also happens with real regex, like:

ucg -r univ.*code.*rep path/to/ucg
/ucg: error: Error during regex parsing: PCRE2 JIT compilation error: bad JIT option

[gvansickle]

Oh, wait:

libpcre2-8 info:
 Version: 10.22 2016-07-29
 JIT support built in?: no
 JIT target architecture: none

No JIT is almost certainly the problem. Does NetBSD have a JIT-enabled PCRE2 available via its package manager or some other reasonable means?

The question of whether to support a no-JIT PCRE/PCRE2 or not has been nagging at me for some time. It's part of the secret sauce that makes ucg "Faster Than Grep"(tm) in its target use cases. Conversely, having a completely non-working ucg benefits nobody....

[0-wiz-0]

I've just built pcre2 with JIT support, and ucg now works fine both with strings and actual regexp.
Perhaps you can improve the error handling in this case :-)

One other issue: on a PaX MPROTECT system, ucg built this way needs to be marked as not MPROTECT safe. On NetBSD this is done by calling 'paxctl +m ucg' after the build is finished. (see http://netbsd.gw.com/cgi-bin/man-cgi?paxctl++NetBSD-current)
On Linux, it's done using 'paxctl -m ucg' (see https://wiki.gentoo.org/wiki/Hardened/PaX_Quickstart#paxctl).
This is a property of the file, so it can be done before installation or after (but take care that the real binary, and not a libtool shell wrapper script, is the target).

[gvansickle]

Hi again @0-wiz-0 ,

I've just built pcre2 with JIT support, and ucg now works fine both with strings and actual regexp.
Perhaps you can improve the error handling in this case :-)

Indeed. I think I've decided that I should require a JIT-enabled PCRE/PCRE2 at configure-time, which would eliminate this as an error.

One other issue: on a PaX MPROTECT system, ucg built this way needs to be marked as not MPROTECT safe.

Yeah, I've been wondering when I was going to run into issues with JIT vs. the NX bit. Thanks for the info; I'll file a separate issue for that.