bsdgrep trivial improvement: define REG_NOSPEC to be a no-op if not defined, use REG_NOSPEC instead of magic number

[kevans]

Summary

At the moment, glibc's cflags are not up to 0020 in value. Future proof this by using REG_NOSPEC directly and defining it to 0 for a no-op in other implementations where it's not defined.

Clearly we can define REG_NOSPEC differently if we pair bsdgrep with another regex(3) implementation that uses a different name, so this is the safe route.

Test Plan

Ensure no regressions with Kyua test suite

[kevans]

• Fix line length

[kevans]

• Apologies, typo

[emaste]

I agree this is better than 0020. But maybe we should have GREP_FIXED be disabled if neither REG_NOSPEC or REG_LITERAL are defined?

Do we have positive and negative test cases for grep / fgrep / egrep already?

[kevans]

In D10420#216375, @emaste wrote:

I agree this is better than 0020. But maybe we should have GREP_FIXED be disabled if neiter REG_NOSPEC or REG_LITERAL are defined?

I considered this, but I've never actually seen an implementation that defined REG_LITERAL -- after reading the commment, I assumed it was a typo since I couldn't find any history of a REG_LITERAL in FreeBSDs regex.h and I hadn't seen an implementation that used it. A quick Google search revealed that I was definitively wrong. =)

Should I use this method of #define REG_LITERAL 0 if not defined and just set cflags |= REG_NOSPEC | REG_LITERAL, or would it be cleaner to have two #ifdef blocks inline in this switch? I personally prefer the former, but if there's a precedent it would be good to follow it.

Do we have positive and negative test cases for grep / fgrep / egrep already?

Heh, well, not really. We don't actually even have a single test that uses fgrep/grep -F. egrep has a primitive test for basic branching functionality/matching escaped characters, but the rest of its inclusion was incidental, having come from the one of the PRs for general matching broken-ness with the -o flag.

We could still use improvement in this area. =)

[kevans]

• Error out at runtime on GREP_FIXED if literal expressions are not supported in regex(3)

[kevans]

Other cruft snuck in despite a recent merge with -HEAD...

[kevans]

Grrr, arc's behavior with Git is not friendly. Diffs against origin/master, not master

[emaste]

usr.bin/grep/grep.c

709

This won't work as DEFINED (upper-case) would be treated as a macro. We want #elif defined(), and I'll make the REG_NOSPEC case use defined() as well, for consistency.

[emaste]

usr.bin/grep/grep.c

712

also we want errx here - there is no errno to report.