preliminary pcre2 support

[broly]

hi,

since my DD-WRT build is now moving to pcre2, i thought i'd give a stab at providing pcre2 support.

since brainslayer is using an ass-old version of MC, i have provided a patch set of the changes necessary to use pcre2.

it compiles fine and from what i can see, it runs fine.

but i haven't tested the search properly because i don't want to restart my router to accommodate the newer glib/pcre (i used LD_LIBRARY_PATH with the newer libraries to test the new executable).

[broly]

if anyone wants to fix any potential bugs that may arise from my changes, use these:

​https://github.com/apache/httpd/commit/c602ba14811ede722017c4e59e4e30d9990227b4#diff-4d479e1bf17e6539c339b182d3c0861a1bd0b0ccbad5226c437d0e30f8cae0c7

​https://github.com/i3/i3/issues/4682#issuecomment-973076704

i will note that, as of the second patch, i have tested the search and it works exactly as you expect!

very happy about this, considering how annoying privoxy is for similar changes.

ENJOY BOYS!

[andrew_b]

The patch is corrupted:

$ git apply shitty_pcre2.patch
shitty_pcre2.patch:27: trailing whitespace.
    lc_mc_search->num_results = pcre2_match (regex, 
shitty_pcre2.patch:53: trailing whitespace.
        int pcre_options = 
shitty_pcre2.patch:56: trailing whitespace.
#else           
shitty_pcre2.patch:62: trailing whitespace.
            pcre_options |= 
shitty_pcre2.patch:70: trailing whitespace.
                pcre_options |= 
error: corrupt patch at line 113

[broly]

Replying to andrew_b:

The patch is corrupted:

$ git apply shitty_pcre2.patch
shitty_pcre2.patch:27: trailing whitespace.
    lc_mc_search->num_results = pcre2_match (regex, 
shitty_pcre2.patch:53: trailing whitespace.
        int pcre_options = 
shitty_pcre2.patch:56: trailing whitespace.
#else           
shitty_pcre2.patch:62: trailing whitespace.
            pcre_options |= 
shitty_pcre2.patch:70: trailing whitespace.
                pcre_options |= 
error: corrupt patch at line 113

sorry for the late reply. turns out trying to jam in a forgotten macro corrupted the patch, because i didn't wanna make a new diff and then delete all of the differences for the */Makefile.in files xD

the updated patch has been tested and applied on my end this time. check it out.

[broly]

fix patch corruption issue.

[andrew_b]

Branch: 4450_pcre2
changeset:49624e473bef82ab56e5e89271f6c4b4cd0f9e5c

@broly it would be nice if you revealed your real name for AUTHORS file.

[andrew_b]

Merged to master: [86a9e0be2cdf01297c902bc4d52dbca778911834].

[eugenesan]

I've been doing some testing with PCRE2 enabled MC and found that it is ~2x faster but provides different search results, which is concerning.

I performed "Find File with Content (Regular expression)" ^.*pcre.*$" on a tree of mixed text, binaries and archives (total ~90k / 3GB).
PCRE2 compiled MC provided ~0.005% (4476 vs 4453) less results when compared to default search engine.
Most of the misses where in binary files (.jar archives to be precise).

[zaytsev]

This is certainly concerning. Did you figure out whether these missed are "real" misses, or the old search is actually wrong?