add --null to usage() forgot in previous commit. reminded by jmc@, thanks!

grep: add --null flag

makes grep print an ASCII NUL byte after the file name to make the
output unambiguous even in the presence of file names with funny
characters.

A previous iteration of this diff was improved by benno@ and tedu@ and
discussed also with sthen@ and deraadt@. deraadt@ also improved the
manpage changes in this version of the diff.

OK deraadt@

Change line counter from int to unsigned long long to reduce overflow.
In case unsigned long long is miraculously still too small add an
additional overflow detection so we stop counting and add a marker to
couter output.

Input on earlier diff guenther@
OK millert

With -R and an implicit ".", don't prepend file paths with "./"

Looks nicer and matches the output of GNU grep.
ok millert@ deraadt@ visa@ miod@

With -R assume that "." was passed instead of printing a warning

Saner default behavior that matches GNU grep. Diff from miod@, support
from espie@, ok visa@ millert@

add --label to usage. thx jmc

two compat features to allow the zstdgrep script to work.
add --label option to prefix the output instead of filename.
allow using - to mean stdin.
ok deraadt

Delete documentation of --max-count, which is merely an alias of
the documented -m. As a rule, we only document long options when
users can't avoid them because they lack a short version.

As suggested by tedu@, as an exception, leave --context documented
because -C is awkward in so far as it takes an optional option
argument, which is fragile and error-prone and hence generally
discouraged, including by POSIX.

Two years ago, kettenis@, deraadt@, and tedu@ all agreed that this
is what should be done, and jmc@ was happy to accept the direction,
but somehow everybody forgot to commit.

convert fgetln to getline. this improves portability and sets a good
better example for other code to follow. in the common case, grep uses
mmap anyway (so no functional change). despite fgetln doing sneaky things
with stdio internals, preliminary analysis by lauri suggests this may
actually reduce the number of allocations.
from Lauri Tirkkonen.

rework grep_open to be more careful about directories.
cleaner, but should be no functional change.
from Lauri Tirkkonen

- add max-count to SYNOPSIS
- list long options with short, where they have an equivalent
- sync usage()
- minor tweaks

Add support for the non-standard grep -m extension.

grep -m num stops after a maximum of num matches are found.
We support -m0 to match GNU behaviour, but we do not allow negative
numbers.

Manpage help from jmc@, OK deraadt@.

fix exit status on pledge(2) error, where it should be >1

OK millert@ deraadt@

Change all tame callers to namechange to pledge(2).

grep only opens files read-only, reads via stdio or other methods, performs
computation, and outputs result to stdout. (note: in the tame model,
malloc is implicit because stdio needs it, and mmap is implicit since
malloc needs it; libz is satisfied by this environment also).

this tame change consists of 1 line, setting "stdio rpath" before
getopt. this protection is fairly strict. grep could be improved
further by computing a wpathlist based on argv, keeping -R in mind.
feel free to take a shot at it.

grep was an early target of capsicum also. know anyone running capsicum grep?

ok doug

use strtonum to parse the number of lines of context.

this provides better error messages.

ok deraadt@ guenther@

Add warning when user specifies -R but no files, like GNU grep.
OK schwarze@ ian@

Don't include limits.h or sys/limits.h since grep.h already does it
for us.

1. They're flags, not counters. Set to one instead of incrementing.
2. The G flag is useless and never checked. Remove it.
ok millert

use reallocarray()

Prefer setvbuf() to setlinebuf() for portability; ok deraadt@

Make option string/struct const (since it is...). I've had this
in my tree for ages.

Fix exit status when there is an error reading a file.
Reported by Jeramey Crawford, fix adapted from FreeBSD.
OK guenther@

add support for -o to only print the match.
reminded this was useful by ajcoutot

add -H (opposite of -h) to always print name. ok deraadt millert

Remove the "fast" grep code if SMALL. This has the side effect of breaking
fgrep -w, but oh well. ok deraadt millert

Fix "grep -e foo -w" crash. The problem lies in 'e' getopt clause
calling add_pattern, which in turn assumes that the getopt phase
has already finished.

OK otto@

remove some non-POSIX standard non-GNU defacto standard options, mostly
relating to symlinks that you can get with find. And a bonus fts flag fix.
With a small tweak by guenther. ok deraadt guenther jmc millert nicm

use calloc() to avoid malloc(n * m) overflows; checked by djm canacar jsg

- Be explicit on command line checking, instead of relying on patterns,
which may be NULL (e.g. -e '').
- let add_pattern() decide how to deal with empty patterns, don't do
magic in read_patterns().

This unbreaks stuff like grep -e '', and makes grep -f <file> more
POSIX compliant. Semantics for grep -f /dev/null (or any other empty
file) may be questionable, but this case isn't specified by POSIX,
and matching nothing at all seems to be sane.

Thanks to otto@, who mentioned potential problems related to the
-x option with the first patch i sent.

ok jaredy@ (some time ago), otto@, millert@

Check length before checking index of len - 1.

OK moritz@.

Allow zero-length patterns with -x so

$ grep -x ""

matches empty lines as reported on misc@ by
Martin Marusak <marusak@fhpv.unipo.sk>.
Initial diff by otto@ with tweaks by me.

ok otto

Break patterns containing newlines into multiple patterns like POSIX says.
Report by Ralf dot Wildenhues at gmx dot de;
testing by jmc@ ok beck@ millert@

delint; remove redundant vars and functions; ok jaredy@

Make the processing of patterns collected from files specified by -f
delayed so options that affect pattern-building (such as -w) can be
applied evenly to all such patterns.

ok and help otto, ok millert

Protect begin and end of word markers added to the pattern when
using the -w option with parentheses, to avoid operators in the
expressions binding to the markers. Compare [[:<:]]foo|bar[[:>:]]
and [[:<:]](foo|bar)[[:>:]]. Problem spotted by aaron@; ok millert@
aaron@ jaredy@

Remove block based mmap optimization. There are newline problems
(PR 3940, 3941) which can be fixed, but if a match starts at the
end of a block and continues into the next block, no match will be
found. Measurements by millert@ showed that the improvements of
this optimization are non-measurable anyway. Diff from Alexander Taler.

ok millert@

various fixes to make this page a bit clearer and hopefully a bit
more helpful;

this includes some ideas/fixes from otto and jared;

spacing

Add a new past path for fgrep that is just a simplified version of
fastcomp. This makes fgrep faster and fixes the -w flag w/ fgrep.
Also remove free_patterns() since calling free right before exit
is silly. Problem noticed by espie@

Implement --line-buffered
ok millert@ tedu@

Fix anchors (^ or $) in -w mode broken by the last commit's -w overhaul.
With this change we pass the updated regress. Tested and OK by ho@

Previously, in -w mode, for each match on a line grep would check
to see if the match was on a word boundary. However, this missed
lines where the first match was not on a word boundary but a
subsequent match was. Problem originally spotted by miod@

We fix this by using the [[:<:]] and [[:>:]] character classes for
the slow path and by checking the word boundaries in grep_search()
for the fast path instead of doing the checks after running
regexec() or grep_search().

With this change, grep passes the new regress tests 15 and 16.
problem originally spotted by espie@.

Sync usage() with SYNOPSIS in grep(1).

ok deraadt@ jmc@

Fix "grep -number" support for multi-digit numbers. At issue is
the fact that optind refers to the *next* argument to be consumed
by getopt(), not the current one. This means we have to keep track
of when we are working with a new argv entry by hand. OK hugh@

When reallocing pattern, use sizeof(*pattern) not sizeof(int).
Fixes a problem on sparc64 where sizeof(int) != sizeof(pointer).
Based on a patch from Brian Poole; tedu@ OK

range-check numeric arguments (-num, -A num, -B num)

grep should exit(2) on error, not exit(1) (1 means no matches found).
deraadt@ OK

knf

Fix parsing of -NUMBER. We now do things a digit at a time and
keep track of what the last char from getopt was as well as the
previous value of optind so we know when a new number has started.

o remove useless cast to int from gzread() call
o maxPatternLen should be size_t since that's what it is compared against
o remove useless casts of NULL to various pointer types

actually do fgrep. -G -F and -E are now mutally exclusive, and override
the program name as expected. ok millert@

strncpy -> memcpy per deraadt suggestion.
also add a note why we can't use strlcpy.

go back to using strncpy. for long patterns, strlcpy reads too much
of a potentially nontermined src.

faster grep for simple patterns. derived from a patch by sean farley.
this makes searching for constant strings much faster by avoiding regex.
ok deraadt@

minor tweaks

use strlcpy, not strncpy. ok deraadt@ millert@

spelling

better detetection/handling of binary files.
make -a do the right thing, doc and implement -U, -I.
add --help and --mmap for compatibility.
some other minor fixes.

some from NetBSD. ok deraadt@

use __progname instead of progname. ok deraadt

fix usage

proper $OpenBSD$

-DNOZ flag to be used by install media for removing z*grep support, if
needed. (and knf)

support z{e,f}grep

freegrep 0.16

Fix format string bug. I believe this grep is currently not used (in
favor of the GNU version), but it gets fixed anyways. ok mickey@

complete grep clone by harding@motd.ca; everything is there, but it is
slow because libc regexp is slow. Anyone out there have the balls to sit
down and optimize libc regexp?

Change line counter from int to unsigned long long to reduce overflow.
In case unsigned long long is miraculously still too small add an
additional overflow detection so we stop counting and add a marker to
couter output.

Input on earlier diff guenther@
OK millert

With -R and an implicit ".", don't prepend file paths with "./"

Looks nicer and matches the output of GNU grep.
ok millert@ deraadt@ visa@ miod@

With -R assume that "." was passed instead of printing a warning

Saner default behavior that matches GNU grep. Diff from miod@, support
from espie@, ok visa@ millert@

add --label to usage. thx jmc

two compat features to allow the zstdgrep script to work.
add --label option to prefix the output instead of filename.
allow using - to mean stdin.
ok deraadt

Delete documentation of --max-count, which is merely an alias of
the documented -m. As a rule, we only document long options when
users can't avoid them because they lack a short version.

As suggested by tedu@, as an exception, leave --context documented
because -C is awkward in so far as it takes an optional option
argument, which is fragile and error-prone and hence generally
discouraged, including by POSIX.

Two years ago, kettenis@, deraadt@, and tedu@ all agreed that this
is what should be done, and jmc@ was happy to accept the direction,
but somehow everybody forgot to commit.

convert fgetln to getline. this improves portability and sets a good
better example for other code to follow. in the common case, grep uses
mmap anyway (so no functional change). despite fgetln doing sneaky things
with stdio internals, preliminary analysis by lauri suggests this may
actually reduce the number of allocations.
from Lauri Tirkkonen.

rework grep_open to be more careful about directories.
cleaner, but should be no functional change.
from Lauri Tirkkonen

- add max-count to SYNOPSIS
- list long options with short, where they have an equivalent
- sync usage()
- minor tweaks

Add support for the non-standard grep -m extension.

grep -m num stops after a maximum of num matches are found.
We support -m0 to match GNU behaviour, but we do not allow negative
numbers.

Manpage help from jmc@, OK deraadt@.

fix exit status on pledge(2) error, where it should be >1

OK millert@ deraadt@

Change all tame callers to namechange to pledge(2).

grep only opens files read-only, reads via stdio or other methods, performs
computation, and outputs result to stdout. (note: in the tame model,
malloc is implicit because stdio needs it, and mmap is implicit since
malloc needs it; libz is satisfied by this environment also).

this tame change consists of 1 line, setting "stdio rpath" before
getopt. this protection is fairly strict. grep could be improved
further by computing a wpathlist based on argv, keeping -R in mind.
feel free to take a shot at it.

grep was an early target of capsicum also. know anyone running capsicum grep?

ok doug

use strtonum to parse the number of lines of context.

this provides better error messages.

ok deraadt@ guenther@

Add warning when user specifies -R but no files, like GNU grep.
OK schwarze@ ian@

Don't include limits.h or sys/limits.h since grep.h already does it
for us.

1. They're flags, not counters. Set to one instead of incrementing.
2. The G flag is useless and never checked. Remove it.
ok millert

use reallocarray()

Prefer setvbuf() to setlinebuf() for portability; ok deraadt@

Make option string/struct const (since it is...). I've had this
in my tree for ages.

Fix exit status when there is an error reading a file.
Reported by Jeramey Crawford, fix adapted from FreeBSD.
OK guenther@

add support for -o to only print the match.
reminded this was useful by ajcoutot

add -H (opposite of -h) to always print name. ok deraadt millert

Remove the "fast" grep code if SMALL. This has the side effect of breaking
fgrep -w, but oh well. ok deraadt millert

Fix "grep -e foo -w" crash. The problem lies in 'e' getopt clause
calling add_pattern, which in turn assumes that the getopt phase
has already finished.

OK otto@

remove some non-POSIX standard non-GNU defacto standard options, mostly
relating to symlinks that you can get with find. And a bonus fts flag fix.
With a small tweak by guenther. ok deraadt guenther jmc millert nicm

use calloc() to avoid malloc(n * m) overflows; checked by djm canacar jsg

- Be explicit on command line checking, instead of relying on patterns,
which may be NULL (e.g. -e '').
- let add_pattern() decide how to deal with empty patterns, don't do
magic in read_patterns().

This unbreaks stuff like grep -e '', and makes grep -f <file> more
POSIX compliant. Semantics for grep -f /dev/null (or any other empty
file) may be questionable, but this case isn't specified by POSIX,
and matching nothing at all seems to be sane.

Thanks to otto@, who mentioned potential problems related to the
-x option with the first patch i sent.

ok jaredy@ (some time ago), otto@, millert@

Check length before checking index of len - 1.

OK moritz@.

Allow zero-length patterns with -x so

$ grep -x ""

matches empty lines as reported on misc@ by
Martin Marusak <marusak@fhpv.unipo.sk>.
Initial diff by otto@ with tweaks by me.

ok otto

Break patterns containing newlines into multiple patterns like POSIX says.
Report by Ralf dot Wildenhues at gmx dot de;
testing by jmc@ ok beck@ millert@

delint; remove redundant vars and functions; ok jaredy@

Make the processing of patterns collected from files specified by -f
delayed so options that affect pattern-building (such as -w) can be
applied evenly to all such patterns.

ok and help otto, ok millert

Protect begin and end of word markers added to the pattern when
using the -w option with parentheses, to avoid operators in the
expressions binding to the markers. Compare [[:<:]]foo|bar[[:>:]]
and [[:<:]](foo|bar)[[:>:]]. Problem spotted by aaron@; ok millert@
aaron@ jaredy@

Remove block based mmap optimization. There are newline problems
(PR 3940, 3941) which can be fixed, but if a match starts at the
end of a block and continues into the next block, no match will be
found. Measurements by millert@ showed that the improvements of
this optimization are non-measurable anyway. Diff from Alexander Taler.

ok millert@

various fixes to make this page a bit clearer and hopefully a bit
more helpful;

this includes some ideas/fixes from otto and jared;

spacing

Add a new past path for fgrep that is just a simplified version of
fastcomp. This makes fgrep faster and fixes the -w flag w/ fgrep.
Also remove free_patterns() since calling free right before exit
is silly. Problem noticed by espie@

Implement --line-buffered
ok millert@ tedu@

Fix anchors (^ or $) in -w mode broken by the last commit's -w overhaul.
With this change we pass the updated regress. Tested and OK by ho@

Previously, in -w mode, for each match on a line grep would check
to see if the match was on a word boundary. However, this missed
lines where the first match was not on a word boundary but a
subsequent match was. Problem originally spotted by miod@

We fix this by using the [[:<:]] and [[:>:]] character classes for
the slow path and by checking the word boundaries in grep_search()
for the fast path instead of doing the checks after running
regexec() or grep_search().

With this change, grep passes the new regress tests 15 and 16.
problem originally spotted by espie@.

Sync usage() with SYNOPSIS in grep(1).

ok deraadt@ jmc@

Fix "grep -number" support for multi-digit numbers. At issue is
the fact that optind refers to the *next* argument to be consumed
by getopt(), not the current one. This means we have to keep track
of when we are working with a new argv entry by hand. OK hugh@

When reallocing pattern, use sizeof(*pattern) not sizeof(int).
Fixes a problem on sparc64 where sizeof(int) != sizeof(pointer).
Based on a patch from Brian Poole; tedu@ OK

range-check numeric arguments (-num, -A num, -B num)

grep should exit(2) on error, not exit(1) (1 means no matches found).
deraadt@ OK

knf

Fix parsing of -NUMBER. We now do things a digit at a time and
keep track of what the last char from getopt was as well as the
previous value of optind so we know when a new number has started.

o remove useless cast to int from gzread() call
o maxPatternLen should be size_t since that's what it is compared against
o remove useless casts of NULL to various pointer types

actually do fgrep. -G -F and -E are now mutally exclusive, and override
the program name as expected. ok millert@

strncpy -> memcpy per deraadt suggestion.
also add a note why we can't use strlcpy.

go back to using strncpy. for long patterns, strlcpy reads too much
of a potentially nontermined src.

faster grep for simple patterns. derived from a patch by sean farley.
this makes searching for constant strings much faster by avoiding regex.
ok deraadt@

minor tweaks

use strlcpy, not strncpy. ok deraadt@ millert@

spelling

better detetection/handling of binary files.
make -a do the right thing, doc and implement -U, -I.
add --help and --mmap for compatibility.
some other minor fixes.

some from NetBSD. ok deraadt@

use __progname instead of progname. ok deraadt

fix usage

proper $OpenBSD$

-DNOZ flag to be used by install media for removing z*grep support, if
needed. (and knf)

support z{e,f}grep

freegrep 0.16

Fix format string bug. I believe this grep is currently not used (in
favor of the GNU version), but it gets fixed anyways. ok mickey@

complete grep clone by harding@motd.ca; everything is there, but it is
slow because libc regexp is slow. Anyone out there have the balls to sit
down and optimize libc regexp?

With -R and an implicit ".", don't prepend file paths with "./"

Looks nicer and matches the output of GNU grep.
ok millert@ deraadt@ visa@ miod@

With -R assume that "." was passed instead of printing a warning

Saner default behavior that matches GNU grep. Diff from miod@, support
from espie@, ok visa@ millert@

add --label to usage. thx jmc

two compat features to allow the zstdgrep script to work.
add --label option to prefix the output instead of filename.
allow using - to mean stdin.
ok deraadt

Delete documentation of --max-count, which is merely an alias of
the documented -m. As a rule, we only document long options when
users can't avoid them because they lack a short version.

As suggested by tedu@, as an exception, leave --context documented
because -C is awkward in so far as it takes an optional option
argument, which is fragile and error-prone and hence generally
discouraged, including by POSIX.

Two years ago, kettenis@, deraadt@, and tedu@ all agreed that this
is what should be done, and jmc@ was happy to accept the direction,
but somehow everybody forgot to commit.

convert fgetln to getline. this improves portability and sets a good
better example for other code to follow. in the common case, grep uses
mmap anyway (so no functional change). despite fgetln doing sneaky things
with stdio internals, preliminary analysis by lauri suggests this may
actually reduce the number of allocations.
from Lauri Tirkkonen.

rework grep_open to be more careful about directories.
cleaner, but should be no functional change.
from Lauri Tirkkonen

- add max-count to SYNOPSIS
- list long options with short, where they have an equivalent
- sync usage()
- minor tweaks

Add support for the non-standard grep -m extension.

grep -m num stops after a maximum of num matches are found.
We support -m0 to match GNU behaviour, but we do not allow negative
numbers.

Manpage help from jmc@, OK deraadt@.

fix exit status on pledge(2) error, where it should be >1

OK millert@ deraadt@

Change all tame callers to namechange to pledge(2).

grep only opens files read-only, reads via stdio or other methods, performs
computation, and outputs result to stdout. (note: in the tame model,
malloc is implicit because stdio needs it, and mmap is implicit since
malloc needs it; libz is satisfied by this environment also).

this tame change consists of 1 line, setting "stdio rpath" before
getopt. this protection is fairly strict. grep could be improved
further by computing a wpathlist based on argv, keeping -R in mind.
feel free to take a shot at it.

grep was an early target of capsicum also. know anyone running capsicum grep?

ok doug

use strtonum to parse the number of lines of context.

this provides better error messages.

ok deraadt@ guenther@

Add warning when user specifies -R but no files, like GNU grep.
OK schwarze@ ian@

Don't include limits.h or sys/limits.h since grep.h already does it
for us.

1. They're flags, not counters. Set to one instead of incrementing.
2. The G flag is useless and never checked. Remove it.
ok millert

use reallocarray()

Prefer setvbuf() to setlinebuf() for portability; ok deraadt@

Make option string/struct const (since it is...). I've had this
in my tree for ages.

Fix exit status when there is an error reading a file.
Reported by Jeramey Crawford, fix adapted from FreeBSD.
OK guenther@

add support for -o to only print the match.
reminded this was useful by ajcoutot

add -H (opposite of -h) to always print name. ok deraadt millert

Remove the "fast" grep code if SMALL. This has the side effect of breaking
fgrep -w, but oh well. ok deraadt millert

Fix "grep -e foo -w" crash. The problem lies in 'e' getopt clause
calling add_pattern, which in turn assumes that the getopt phase
has already finished.

OK otto@

remove some non-POSIX standard non-GNU defacto standard options, mostly
relating to symlinks that you can get with find. And a bonus fts flag fix.
With a small tweak by guenther. ok deraadt guenther jmc millert nicm

use calloc() to avoid malloc(n * m) overflows; checked by djm canacar jsg

- Be explicit on command line checking, instead of relying on patterns,
which may be NULL (e.g. -e '').
- let add_pattern() decide how to deal with empty patterns, don't do
magic in read_patterns().

This unbreaks stuff like grep -e '', and makes grep -f <file> more
POSIX compliant. Semantics for grep -f /dev/null (or any other empty
file) may be questionable, but this case isn't specified by POSIX,
and matching nothing at all seems to be sane.

Thanks to otto@, who mentioned potential problems related to the
-x option with the first patch i sent.

ok jaredy@ (some time ago), otto@, millert@

Check length before checking index of len - 1.

OK moritz@.

Allow zero-length patterns with -x so

$ grep -x ""

matches empty lines as reported on misc@ by
Martin Marusak <marusak@fhpv.unipo.sk>.
Initial diff by otto@ with tweaks by me.

ok otto

Break patterns containing newlines into multiple patterns like POSIX says.
Report by Ralf dot Wildenhues at gmx dot de;
testing by jmc@ ok beck@ millert@

delint; remove redundant vars and functions; ok jaredy@

Make the processing of patterns collected from files specified by -f
delayed so options that affect pattern-building (such as -w) can be
applied evenly to all such patterns.

ok and help otto, ok millert

Protect begin and end of word markers added to the pattern when
using the -w option with parentheses, to avoid operators in the
expressions binding to the markers. Compare [[:<:]]foo|bar[[:>:]]
and [[:<:]](foo|bar)[[:>:]]. Problem spotted by aaron@; ok millert@
aaron@ jaredy@

Remove block based mmap optimization. There are newline problems
(PR 3940, 3941) which can be fixed, but if a match starts at the
end of a block and continues into the next block, no match will be
found. Measurements by millert@ showed that the improvements of
this optimization are non-measurable anyway. Diff from Alexander Taler.

ok millert@

various fixes to make this page a bit clearer and hopefully a bit
more helpful;

this includes some ideas/fixes from otto and jared;

spacing

Add a new past path for fgrep that is just a simplified version of
fastcomp. This makes fgrep faster and fixes the -w flag w/ fgrep.
Also remove free_patterns() since calling free right before exit
is silly. Problem noticed by espie@

Implement --line-buffered
ok millert@ tedu@

Fix anchors (^ or $) in -w mode broken by the last commit's -w overhaul.
With this change we pass the updated regress. Tested and OK by ho@

Previously, in -w mode, for each match on a line grep would check
to see if the match was on a word boundary. However, this missed
lines where the first match was not on a word boundary but a
subsequent match was. Problem originally spotted by miod@

We fix this by using the [[:<:]] and [[:>:]] character classes for
the slow path and by checking the word boundaries in grep_search()
for the fast path instead of doing the checks after running
regexec() or grep_search().

With this change, grep passes the new regress tests 15 and 16.
problem originally spotted by espie@.

Sync usage() with SYNOPSIS in grep(1).

ok deraadt@ jmc@

Fix "grep -number" support for multi-digit numbers. At issue is
the fact that optind refers to the *next* argument to be consumed
by getopt(), not the current one. This means we have to keep track
of when we are working with a new argv entry by hand. OK hugh@

When reallocing pattern, use sizeof(*pattern) not sizeof(int).
Fixes a problem on sparc64 where sizeof(int) != sizeof(pointer).
Based on a patch from Brian Poole; tedu@ OK

range-check numeric arguments (-num, -A num, -B num)

grep should exit(2) on error, not exit(1) (1 means no matches found).
deraadt@ OK

knf

Fix parsing of -NUMBER. We now do things a digit at a time and
keep track of what the last char from getopt was as well as the
previous value of optind so we know when a new number has started.

o remove useless cast to int from gzread() call
o maxPatternLen should be size_t since that's what it is compared against
o remove useless casts of NULL to various pointer types

actually do fgrep. -G -F and -E are now mutally exclusive, and override
the program name as expected. ok millert@

strncpy -> memcpy per deraadt suggestion.
also add a note why we can't use strlcpy.

go back to using strncpy. for long patterns, strlcpy reads too much
of a potentially nontermined src.

faster grep for simple patterns. derived from a patch by sean farley.
this makes searching for constant strings much faster by avoiding regex.
ok deraadt@

minor tweaks

use strlcpy, not strncpy. ok deraadt@ millert@

spelling

better detetection/handling of binary files.
make -a do the right thing, doc and implement -U, -I.
add --help and --mmap for compatibility.
some other minor fixes.

some from NetBSD. ok deraadt@

use __progname instead of progname. ok deraadt

fix usage

proper $OpenBSD$

-DNOZ flag to be used by install media for removing z*grep support, if
needed. (and knf)

support z{e,f}grep

freegrep 0.16

Fix format string bug. I believe this grep is currently not used (in
favor of the GNU version), but it gets fixed anyways. ok mickey@

complete grep clone by harding@motd.ca; everything is there, but it is
slow because libc regexp is slow. Anyone out there have the balls to sit
down and optimize libc regexp?

add --label to usage. thx jmc

two compat features to allow the zstdgrep script to work.
add --label option to prefix the output instead of filename.
allow using - to mean stdin.
ok deraadt

Delete documentation of --max-count, which is merely an alias of
the documented -m. As a rule, we only document long options when
users can't avoid them because they lack a short version.

As suggested by tedu@, as an exception, leave --context documented
because -C is awkward in so far as it takes an optional option
argument, which is fragile and error-prone and hence generally
discouraged, including by POSIX.

Two years ago, kettenis@, deraadt@, and tedu@ all agreed that this
is what should be done, and jmc@ was happy to accept the direction,
but somehow everybody forgot to commit.

convert fgetln to getline. this improves portability and sets a good
better example for other code to follow. in the common case, grep uses
mmap anyway (so no functional change). despite fgetln doing sneaky things
with stdio internals, preliminary analysis by lauri suggests this may
actually reduce the number of allocations.
from Lauri Tirkkonen.

rework grep_open to be more careful about directories.
cleaner, but should be no functional change.
from Lauri Tirkkonen

- add max-count to SYNOPSIS
- list long options with short, where they have an equivalent
- sync usage()
- minor tweaks

Add support for the non-standard grep -m extension.

grep -m num stops after a maximum of num matches are found.
We support -m0 to match GNU behaviour, but we do not allow negative
numbers.

Manpage help from jmc@, OK deraadt@.

fix exit status on pledge(2) error, where it should be >1

OK millert@ deraadt@

Change all tame callers to namechange to pledge(2).

grep only opens files read-only, reads via stdio or other methods, performs
computation, and outputs result to stdout. (note: in the tame model,
malloc is implicit because stdio needs it, and mmap is implicit since
malloc needs it; libz is satisfied by this environment also).

this tame change consists of 1 line, setting "stdio rpath" before
getopt. this protection is fairly strict. grep could be improved
further by computing a wpathlist based on argv, keeping -R in mind.
feel free to take a shot at it.

grep was an early target of capsicum also. know anyone running capsicum grep?

ok doug

use strtonum to parse the number of lines of context.

this provides better error messages.

ok deraadt@ guenther@

Add warning when user specifies -R but no files, like GNU grep.
OK schwarze@ ian@

Don't include limits.h or sys/limits.h since grep.h already does it
for us.

1. They're flags, not counters. Set to one instead of incrementing.
2. The G flag is useless and never checked. Remove it.
ok millert

use reallocarray()

Prefer setvbuf() to setlinebuf() for portability; ok deraadt@

Make option string/struct const (since it is...). I've had this
in my tree for ages.

Fix exit status when there is an error reading a file.
Reported by Jeramey Crawford, fix adapted from FreeBSD.
OK guenther@

add support for -o to only print the match.
reminded this was useful by ajcoutot

add -H (opposite of -h) to always print name. ok deraadt millert

Remove the "fast" grep code if SMALL. This has the side effect of breaking
fgrep -w, but oh well. ok deraadt millert

Fix "grep -e foo -w" crash. The problem lies in 'e' getopt clause
calling add_pattern, which in turn assumes that the getopt phase
has already finished.

OK otto@

remove some non-POSIX standard non-GNU defacto standard options, mostly
relating to symlinks that you can get with find. And a bonus fts flag fix.
With a small tweak by guenther. ok deraadt guenther jmc millert nicm

use calloc() to avoid malloc(n * m) overflows; checked by djm canacar jsg

- Be explicit on command line checking, instead of relying on patterns,
which may be NULL (e.g. -e '').
- let add_pattern() decide how to deal with empty patterns, don't do
magic in read_patterns().

This unbreaks stuff like grep -e '', and makes grep -f <file> more
POSIX compliant. Semantics for grep -f /dev/null (or any other empty
file) may be questionable, but this case isn't specified by POSIX,
and matching nothing at all seems to be sane.

Thanks to otto@, who mentioned potential problems related to the
-x option with the first patch i sent.

ok jaredy@ (some time ago), otto@, millert@

Check length before checking index of len - 1.

OK moritz@.

Allow zero-length patterns with -x so

$ grep -x ""

matches empty lines as reported on misc@ by
Martin Marusak <marusak@fhpv.unipo.sk>.
Initial diff by otto@ with tweaks by me.

ok otto

Break patterns containing newlines into multiple patterns like POSIX says.
Report by Ralf dot Wildenhues at gmx dot de;
testing by jmc@ ok beck@ millert@

delint; remove redundant vars and functions; ok jaredy@

Make the processing of patterns collected from files specified by -f
delayed so options that affect pattern-building (such as -w) can be
applied evenly to all such patterns.

ok and help otto, ok millert

Protect begin and end of word markers added to the pattern when
using the -w option with parentheses, to avoid operators in the
expressions binding to the markers. Compare [[:<:]]foo|bar[[:>:]]
and [[:<:]](foo|bar)[[:>:]]. Problem spotted by aaron@; ok millert@
aaron@ jaredy@

Remove block based mmap optimization. There are newline problems
(PR 3940, 3941) which can be fixed, but if a match starts at the
end of a block and continues into the next block, no match will be
found. Measurements by millert@ showed that the improvements of
this optimization are non-measurable anyway. Diff from Alexander Taler.

ok millert@

various fixes to make this page a bit clearer and hopefully a bit
more helpful;

this includes some ideas/fixes from otto and jared;

spacing

Add a new past path for fgrep that is just a simplified version of
fastcomp. This makes fgrep faster and fixes the -w flag w/ fgrep.
Also remove free_patterns() since calling free right before exit
is silly. Problem noticed by espie@

Implement --line-buffered
ok millert@ tedu@

Fix anchors (^ or $) in -w mode broken by the last commit's -w overhaul.
With this change we pass the updated regress. Tested and OK by ho@

Previously, in -w mode, for each match on a line grep would check
to see if the match was on a word boundary. However, this missed
lines where the first match was not on a word boundary but a
subsequent match was. Problem originally spotted by miod@

We fix this by using the [[:<:]] and [[:>:]] character classes for
the slow path and by checking the word boundaries in grep_search()
for the fast path instead of doing the checks after running
regexec() or grep_search().

With this change, grep passes the new regress tests 15 and 16.
problem originally spotted by espie@.

Sync usage() with SYNOPSIS in grep(1).

ok deraadt@ jmc@

Fix "grep -number" support for multi-digit numbers. At issue is
the fact that optind refers to the *next* argument to be consumed
by getopt(), not the current one. This means we have to keep track
of when we are working with a new argv entry by hand. OK hugh@

When reallocing pattern, use sizeof(*pattern) not sizeof(int).
Fixes a problem on sparc64 where sizeof(int) != sizeof(pointer).
Based on a patch from Brian Poole; tedu@ OK

range-check numeric arguments (-num, -A num, -B num)

grep should exit(2) on error, not exit(1) (1 means no matches found).
deraadt@ OK

knf

Fix parsing of -NUMBER. We now do things a digit at a time and
keep track of what the last char from getopt was as well as the
previous value of optind so we know when a new number has started.

o remove useless cast to int from gzread() call
o maxPatternLen should be size_t since that's what it is compared against
o remove useless casts of NULL to various pointer types

actually do fgrep. -G -F and -E are now mutally exclusive, and override
the program name as expected. ok millert@

strncpy -> memcpy per deraadt suggestion.
also add a note why we can't use strlcpy.

go back to using strncpy. for long patterns, strlcpy reads too much
of a potentially nontermined src.

faster grep for simple patterns. derived from a patch by sean farley.
this makes searching for constant strings much faster by avoiding regex.
ok deraadt@

minor tweaks

use strlcpy, not strncpy. ok deraadt@ millert@

spelling

better detetection/handling of binary files.
make -a do the right thing, doc and implement -U, -I.
add --help and --mmap for compatibility.
some other minor fixes.

some from NetBSD. ok deraadt@

use __progname instead of progname. ok deraadt

fix usage

proper $OpenBSD$

-DNOZ flag to be used by install media for removing z*grep support, if
needed. (and knf)

support z{e,f}grep

freegrep 0.16

Fix format string bug. I believe this grep is currently not used (in
favor of the GNU version), but it gets fixed anyways. ok mickey@

complete grep clone by harding@motd.ca; everything is there, but it is
slow because libc regexp is slow. Anyone out there have the balls to sit
down and optimize libc regexp?

Delete documentation of --max-count, which is merely an alias of
the documented -m. As a rule, we only document long options when
users can't avoid them because they lack a short version.

As suggested by tedu@, as an exception, leave --context documented
because -C is awkward in so far as it takes an optional option
argument, which is fragile and error-prone and hence generally
discouraged, including by POSIX.

Two years ago, kettenis@, deraadt@, and tedu@ all agreed that this
is what should be done, and jmc@ was happy to accept the direction,
but somehow everybody forgot to commit.

convert fgetln to getline. this improves portability and sets a good
better example for other code to follow. in the common case, grep uses
mmap anyway (so no functional change). despite fgetln doing sneaky things
with stdio internals, preliminary analysis by lauri suggests this may
actually reduce the number of allocations.
from Lauri Tirkkonen.

rework grep_open to be more careful about directories.
cleaner, but should be no functional change.
from Lauri Tirkkonen

- add max-count to SYNOPSIS
- list long options with short, where they have an equivalent
- sync usage()
- minor tweaks

Add support for the non-standard grep -m extension.

grep -m num stops after a maximum of num matches are found.
We support -m0 to match GNU behaviour, but we do not allow negative
numbers.

Manpage help from jmc@, OK deraadt@.

fix exit status on pledge(2) error, where it should be >1

OK millert@ deraadt@

Change all tame callers to namechange to pledge(2).

grep only opens files read-only, reads via stdio or other methods, performs
computation, and outputs result to stdout. (note: in the tame model,
malloc is implicit because stdio needs it, and mmap is implicit since
malloc needs it; libz is satisfied by this environment also).

this tame change consists of 1 line, setting "stdio rpath" before
getopt. this protection is fairly strict. grep could be improved
further by computing a wpathlist based on argv, keeping -R in mind.
feel free to take a shot at it.

grep was an early target of capsicum also. know anyone running capsicum grep?

ok doug

use strtonum to parse the number of lines of context.

this provides better error messages.

ok deraadt@ guenther@

Add warning when user specifies -R but no files, like GNU grep.
OK schwarze@ ian@

Don't include limits.h or sys/limits.h since grep.h already does it
for us.

1. They're flags, not counters. Set to one instead of incrementing.
2. The G flag is useless and never checked. Remove it.
ok millert

use reallocarray()

Prefer setvbuf() to setlinebuf() for portability; ok deraadt@

Make option string/struct const (since it is...). I've had this
in my tree for ages.

Fix exit status when there is an error reading a file.
Reported by Jeramey Crawford, fix adapted from FreeBSD.
OK guenther@

add support for -o to only print the match.
reminded this was useful by ajcoutot

add -H (opposite of -h) to always print name. ok deraadt millert

Remove the "fast" grep code if SMALL. This has the side effect of breaking
fgrep -w, but oh well. ok deraadt millert

Fix "grep -e foo -w" crash. The problem lies in 'e' getopt clause
calling add_pattern, which in turn assumes that the getopt phase
has already finished.

OK otto@

remove some non-POSIX standard non-GNU defacto standard options, mostly
relating to symlinks that you can get with find. And a bonus fts flag fix.
With a small tweak by guenther. ok deraadt guenther jmc millert nicm

use calloc() to avoid malloc(n * m) overflows; checked by djm canacar jsg

- Be explicit on command line checking, instead of relying on patterns,
which may be NULL (e.g. -e '').
- let add_pattern() decide how to deal with empty patterns, don't do
magic in read_patterns().

This unbreaks stuff like grep -e '', and makes grep -f <file> more
POSIX compliant. Semantics for grep -f /dev/null (or any other empty
file) may be questionable, but this case isn't specified by POSIX,
and matching nothing at all seems to be sane.

Thanks to otto@, who mentioned potential problems related to the
-x option with the first patch i sent.

ok jaredy@ (some time ago), otto@, millert@

Check length before checking index of len - 1.

OK moritz@.

Allow zero-length patterns with -x so

$ grep -x ""

matches empty lines as reported on misc@ by
Martin Marusak <marusak@fhpv.unipo.sk>.
Initial diff by otto@ with tweaks by me.

ok otto

Break patterns containing newlines into multiple patterns like POSIX says.
Report by Ralf dot Wildenhues at gmx dot de;
testing by jmc@ ok beck@ millert@

delint; remove redundant vars and functions; ok jaredy@

Make the processing of patterns collected from files specified by -f
delayed so options that affect pattern-building (such as -w) can be
applied evenly to all such patterns.

ok and help otto, ok millert

Protect begin and end of word markers added to the pattern when
using the -w option with parentheses, to avoid operators in the
expressions binding to the markers. Compare [[:<:]]foo|bar[[:>:]]
and [[:<:]](foo|bar)[[:>:]]. Problem spotted by aaron@; ok millert@
aaron@ jaredy@

Remove block based mmap optimization. There are newline problems
(PR 3940, 3941) which can be fixed, but if a match starts at the
end of a block and continues into the next block, no match will be
found. Measurements by millert@ showed that the improvements of
this optimization are non-measurable anyway. Diff from Alexander Taler.

ok millert@

various fixes to make this page a bit clearer and hopefully a bit
more helpful;

this includes some ideas/fixes from otto and jared;

spacing

Add a new past path for fgrep that is just a simplified version of
fastcomp. This makes fgrep faster and fixes the -w flag w/ fgrep.
Also remove free_patterns() since calling free right before exit
is silly. Problem noticed by espie@

Implement --line-buffered
ok millert@ tedu@

Fix anchors (^ or $) in -w mode broken by the last commit's -w overhaul.
With this change we pass the updated regress. Tested and OK by ho@

Previously, in -w mode, for each match on a line grep would check
to see if the match was on a word boundary. However, this missed
lines where the first match was not on a word boundary but a
subsequent match was. Problem originally spotted by miod@

We fix this by using the [[:<:]] and [[:>:]] character classes for
the slow path and by checking the word boundaries in grep_search()
for the fast path instead of doing the checks after running
regexec() or grep_search().

With this change, grep passes the new regress tests 15 and 16.
problem originally spotted by espie@.

Sync usage() with SYNOPSIS in grep(1).

ok deraadt@ jmc@

Fix "grep -number" support for multi-digit numbers. At issue is
the fact that optind refers to the *next* argument to be consumed
by getopt(), not the current one. This means we have to keep track
of when we are working with a new argv entry by hand. OK hugh@

When reallocing pattern, use sizeof(*pattern) not sizeof(int).
Fixes a problem on sparc64 where sizeof(int) != sizeof(pointer).
Based on a patch from Brian Poole; tedu@ OK

range-check numeric arguments (-num, -A num, -B num)

grep should exit(2) on error, not exit(1) (1 means no matches found).
deraadt@ OK

knf

Fix parsing of -NUMBER. We now do things a digit at a time and
keep track of what the last char from getopt was as well as the
previous value of optind so we know when a new number has started.

o remove useless cast to int from gzread() call
o maxPatternLen should be size_t since that's what it is compared against
o remove useless casts of NULL to various pointer types

actually do fgrep. -G -F and -E are now mutally exclusive, and override
the program name as expected. ok millert@

strncpy -> memcpy per deraadt suggestion.
also add a note why we can't use strlcpy.

go back to using strncpy. for long patterns, strlcpy reads too much
of a potentially nontermined src.

faster grep for simple patterns. derived from a patch by sean farley.
this makes searching for constant strings much faster by avoiding regex.
ok deraadt@

minor tweaks

use strlcpy, not strncpy. ok deraadt@ millert@

spelling

better detetection/handling of binary files.
make -a do the right thing, doc and implement -U, -I.
add --help and --mmap for compatibility.
some other minor fixes.

some from NetBSD. ok deraadt@

use __progname instead of progname. ok deraadt

fix usage

proper $OpenBSD$

-DNOZ flag to be used by install media for removing z*grep support, if
needed. (and knf)

support z{e,f}grep

freegrep 0.16

Fix format string bug. I believe this grep is currently not used (in
favor of the GNU version), but it gets fixed anyways. ok mickey@

complete grep clone by harding@motd.ca; everything is there, but it is
slow because libc regexp is slow. Anyone out there have the balls to sit
down and optimize libc regexp?

convert fgetln to getline. this improves portability and sets a good
better example for other code to follow. in the common case, grep uses
mmap anyway (so no functional change). despite fgetln doing sneaky things
with stdio internals, preliminary analysis by lauri suggests this may
actually reduce the number of allocations.
from Lauri Tirkkonen.

rework grep_open to be more careful about directories.
cleaner, but should be no functional change.
from Lauri Tirkkonen

- add max-count to SYNOPSIS
- list long options with short, where they have an equivalent
- sync usage()
- minor tweaks

Add support for the non-standard grep -m extension.

grep -m num stops after a maximum of num matches are found.
We support -m0 to match GNU behaviour, but we do not allow negative
numbers.

Manpage help from jmc@, OK deraadt@.

fix exit status on pledge(2) error, where it should be >1

OK millert@ deraadt@

Change all tame callers to namechange to pledge(2).

grep only opens files read-only, reads via stdio or other methods, performs
computation, and outputs result to stdout. (note: in the tame model,
malloc is implicit because stdio needs it, and mmap is implicit since
malloc needs it; libz is satisfied by this environment also).

this tame change consists of 1 line, setting "stdio rpath" before
getopt. this protection is fairly strict. grep could be improved
further by computing a wpathlist based on argv, keeping -R in mind.
feel free to take a shot at it.

grep was an early target of capsicum also. know anyone running capsicum grep?

ok doug

use strtonum to parse the number of lines of context.

this provides better error messages.

ok deraadt@ guenther@

Add warning when user specifies -R but no files, like GNU grep.
OK schwarze@ ian@

Don't include limits.h or sys/limits.h since grep.h already does it
for us.

1. They're flags, not counters. Set to one instead of incrementing.
2. The G flag is useless and never checked. Remove it.
ok millert

use reallocarray()

Prefer setvbuf() to setlinebuf() for portability; ok deraadt@

Make option string/struct const (since it is...). I've had this
in my tree for ages.

Fix exit status when there is an error reading a file.
Reported by Jeramey Crawford, fix adapted from FreeBSD.
OK guenther@

add support for -o to only print the match.
reminded this was useful by ajcoutot

add -H (opposite of -h) to always print name. ok deraadt millert

Remove the "fast" grep code if SMALL. This has the side effect of breaking
fgrep -w, but oh well. ok deraadt millert

Fix "grep -e foo -w" crash. The problem lies in 'e' getopt clause
calling add_pattern, which in turn assumes that the getopt phase
has already finished.

OK otto@

remove some non-POSIX standard non-GNU defacto standard options, mostly
relating to symlinks that you can get with find. And a bonus fts flag fix.
With a small tweak by guenther. ok deraadt guenther jmc millert nicm

use calloc() to avoid malloc(n * m) overflows; checked by djm canacar jsg

- Be explicit on command line checking, instead of relying on patterns,
which may be NULL (e.g. -e '').
- let add_pattern() decide how to deal with empty patterns, don't do
magic in read_patterns().

This unbreaks stuff like grep -e '', and makes grep -f <file> more
POSIX compliant. Semantics for grep -f /dev/null (or any other empty
file) may be questionable, but this case isn't specified by POSIX,
and matching nothing at all seems to be sane.

Thanks to otto@, who mentioned potential problems related to the
-x option with the first patch i sent.

ok jaredy@ (some time ago), otto@, millert@

Check length before checking index of len - 1.

OK moritz@.

Allow zero-length patterns with -x so

$ grep -x ""

matches empty lines as reported on misc@ by
Martin Marusak <marusak@fhpv.unipo.sk>.
Initial diff by otto@ with tweaks by me.

ok otto

Break patterns containing newlines into multiple patterns like POSIX says.
Report by Ralf dot Wildenhues at gmx dot de;
testing by jmc@ ok beck@ millert@

delint; remove redundant vars and functions; ok jaredy@

Make the processing of patterns collected from files specified by -f
delayed so options that affect pattern-building (such as -w) can be
applied evenly to all such patterns.

ok and help otto, ok millert

Protect begin and end of word markers added to the pattern when
using the -w option with parentheses, to avoid operators in the
expressions binding to the markers. Compare [[:<:]]foo|bar[[:>:]]
and [[:<:]](foo|bar)[[:>:]]. Problem spotted by aaron@; ok millert@
aaron@ jaredy@

Remove block based mmap optimization. There are newline problems
(PR 3940, 3941) which can be fixed, but if a match starts at the
end of a block and continues into the next block, no match will be
found. Measurements by millert@ showed that the improvements of
this optimization are non-measurable anyway. Diff from Alexander Taler.

ok millert@

various fixes to make this page a bit clearer and hopefully a bit
more helpful;

this includes some ideas/fixes from otto and jared;

spacing

Add a new past path for fgrep that is just a simplified version of
fastcomp. This makes fgrep faster and fixes the -w flag w/ fgrep.
Also remove free_patterns() since calling free right before exit
is silly. Problem noticed by espie@

Implement --line-buffered
ok millert@ tedu@

Fix anchors (^ or $) in -w mode broken by the last commit's -w overhaul.
With this change we pass the updated regress. Tested and OK by ho@

Previously, in -w mode, for each match on a line grep would check
to see if the match was on a word boundary. However, this missed
lines where the first match was not on a word boundary but a
subsequent match was. Problem originally spotted by miod@

We fix this by using the [[:<:]] and [[:>:]] character classes for
the slow path and by checking the word boundaries in grep_search()
for the fast path instead of doing the checks after running
regexec() or grep_search().

With this change, grep passes the new regress tests 15 and 16.
problem originally spotted by espie@.

Sync usage() with SYNOPSIS in grep(1).

ok deraadt@ jmc@

Fix "grep -number" support for multi-digit numbers. At issue is
the fact that optind refers to the *next* argument to be consumed
by getopt(), not the current one. This means we have to keep track
of when we are working with a new argv entry by hand. OK hugh@

When reallocing pattern, use sizeof(*pattern) not sizeof(int).
Fixes a problem on sparc64 where sizeof(int) != sizeof(pointer).
Based on a patch from Brian Poole; tedu@ OK

range-check numeric arguments (-num, -A num, -B num)

grep should exit(2) on error, not exit(1) (1 means no matches found).
deraadt@ OK

knf

Fix parsing of -NUMBER. We now do things a digit at a time and
keep track of what the last char from getopt was as well as the
previous value of optind so we know when a new number has started.

o remove useless cast to int from gzread() call
o maxPatternLen should be size_t since that's what it is compared against
o remove useless casts of NULL to various pointer types

actually do fgrep. -G -F and -E are now mutally exclusive, and override
the program name as expected. ok millert@

strncpy -> memcpy per deraadt suggestion.
also add a note why we can't use strlcpy.

go back to using strncpy. for long patterns, strlcpy reads too much
of a potentially nontermined src.

faster grep for simple patterns. derived from a patch by sean farley.
this makes searching for constant strings much faster by avoiding regex.
ok deraadt@

minor tweaks

use strlcpy, not strncpy. ok deraadt@ millert@

spelling

better detetection/handling of binary files.
make -a do the right thing, doc and implement -U, -I.
add --help and --mmap for compatibility.
some other minor fixes.

some from NetBSD. ok deraadt@

use __progname instead of progname. ok deraadt

fix usage

proper $OpenBSD$

-DNOZ flag to be used by install media for removing z*grep support, if
needed. (and knf)

support z{e,f}grep

freegrep 0.16

Fix format string bug. I believe this grep is currently not used (in
favor of the GNU version), but it gets fixed anyways. ok mickey@

complete grep clone by harding@motd.ca; everything is there, but it is
slow because libc regexp is slow. Anyone out there have the balls to sit
down and optimize libc regexp?

rework grep_open to be more careful about directories.
cleaner, but should be no functional change.
from Lauri Tirkkonen

- add max-count to SYNOPSIS
- list long options with short, where they have an equivalent
- sync usage()
- minor tweaks

Add support for the non-standard grep -m extension.

grep -m num stops after a maximum of num matches are found.
We support -m0 to match GNU behaviour, but we do not allow negative
numbers.

Manpage help from jmc@, OK deraadt@.

fix exit status on pledge(2) error, where it should be >1

OK millert@ deraadt@

Change all tame callers to namechange to pledge(2).

grep only opens files read-only, reads via stdio or other methods, performs
computation, and outputs result to stdout. (note: in the tame model,
malloc is implicit because stdio needs it, and mmap is implicit since
malloc needs it; libz is satisfied by this environment also).

this tame change consists of 1 line, setting "stdio rpath" before
getopt. this protection is fairly strict. grep could be improved
further by computing a wpathlist based on argv, keeping -R in mind.
feel free to take a shot at it.

grep was an early target of capsicum also. know anyone running capsicum grep?

ok doug

use strtonum to parse the number of lines of context.

this provides better error messages.

ok deraadt@ guenther@

Add warning when user specifies -R but no files, like GNU grep.
OK schwarze@ ian@

Don't include limits.h or sys/limits.h since grep.h already does it
for us.

1. They're flags, not counters. Set to one instead of incrementing.
2. The G flag is useless and never checked. Remove it.
ok millert

use reallocarray()

Prefer setvbuf() to setlinebuf() for portability; ok deraadt@

Make option string/struct const (since it is...). I've had this
in my tree for ages.

Fix exit status when there is an error reading a file.
Reported by Jeramey Crawford, fix adapted from FreeBSD.
OK guenther@

add support for -o to only print the match.
reminded this was useful by ajcoutot

add -H (opposite of -h) to always print name. ok deraadt millert

Remove the "fast" grep code if SMALL. This has the side effect of breaking
fgrep -w, but oh well. ok deraadt millert

Fix "grep -e foo -w" crash. The problem lies in 'e' getopt clause
calling add_pattern, which in turn assumes that the getopt phase
has already finished.

OK otto@

remove some non-POSIX standard non-GNU defacto standard options, mostly
relating to symlinks that you can get with find. And a bonus fts flag fix.
With a small tweak by guenther. ok deraadt guenther jmc millert nicm

use calloc() to avoid malloc(n * m) overflows; checked by djm canacar jsg

- Be explicit on command line checking, instead of relying on patterns,
which may be NULL (e.g. -e '').
- let add_pattern() decide how to deal with empty patterns, don't do
magic in read_patterns().

This unbreaks stuff like grep -e '', and makes grep -f <file> more
POSIX compliant. Semantics for grep -f /dev/null (or any other empty
file) may be questionable, but this case isn't specified by POSIX,
and matching nothing at all seems to be sane.

Thanks to otto@, who mentioned potential problems related to the
-x option with the first patch i sent.

ok jaredy@ (some time ago), otto@, millert@

Check length before checking index of len - 1.

OK moritz@.

Allow zero-length patterns with -x so

$ grep -x ""

matches empty lines as reported on misc@ by
Martin Marusak <marusak@fhpv.unipo.sk>.
Initial diff by otto@ with tweaks by me.

ok otto

Break patterns containing newlines into multiple patterns like POSIX says.
Report by Ralf dot Wildenhues at gmx dot de;
testing by jmc@ ok beck@ millert@

delint; remove redundant vars and functions; ok jaredy@

Make the processing of patterns collected from files specified by -f
delayed so options that affect pattern-building (such as -w) can be
applied evenly to all such patterns.

ok and help otto, ok millert

Protect begin and end of word markers added to the pattern when
using the -w option with parentheses, to avoid operators in the
expressions binding to the markers. Compare [[:<:]]foo|bar[[:>:]]
and [[:<:]](foo|bar)[[:>:]]. Problem spotted by aaron@; ok millert@
aaron@ jaredy@

Remove block based mmap optimization. There are newline problems
(PR 3940, 3941) which can be fixed, but if a match starts at the
end of a block and continues into the next block, no match will be
found. Measurements by millert@ showed that the improvements of
this optimization are non-measurable anyway. Diff from Alexander Taler.

ok millert@

various fixes to make this page a bit clearer and hopefully a bit
more helpful;

this includes some ideas/fixes from otto and jared;

spacing

Add a new past path for fgrep that is just a simplified version of
fastcomp. This makes fgrep faster and fixes the -w flag w/ fgrep.
Also remove free_patterns() since calling free right before exit
is silly. Problem noticed by espie@

Implement --line-buffered
ok millert@ tedu@

Fix anchors (^ or $) in -w mode broken by the last commit's -w overhaul.
With this change we pass the updated regress. Tested and OK by ho@

Previously, in -w mode, for each match on a line grep would check
to see if the match was on a word boundary. However, this missed
lines where the first match was not on a word boundary but a
subsequent match was. Problem originally spotted by miod@

We fix this by using the [[:<:]] and [[:>:]] character classes for
the slow path and by checking the word boundaries in grep_search()
for the fast path instead of doing the checks after running
regexec() or grep_search().

With this change, grep passes the new regress tests 15 and 16.
problem originally spotted by espie@.

Sync usage() with SYNOPSIS in grep(1).

ok deraadt@ jmc@

Fix "grep -number" support for multi-digit numbers. At issue is
the fact that optind refers to the *next* argument to be consumed
by getopt(), not the current one. This means we have to keep track
of when we are working with a new argv entry by hand. OK hugh@

When reallocing pattern, use sizeof(*pattern) not sizeof(int).
Fixes a problem on sparc64 where sizeof(int) != sizeof(pointer).
Based on a patch from Brian Poole; tedu@ OK

range-check numeric arguments (-num, -A num, -B num)

grep should exit(2) on error, not exit(1) (1 means no matches found).
deraadt@ OK

knf

Fix parsing of -NUMBER. We now do things a digit at a time and
keep track of what the last char from getopt was as well as the
previous value of optind so we know when a new number has started.

o remove useless cast to int from gzread() call
o maxPatternLen should be size_t since that's what it is compared against
o remove useless casts of NULL to various pointer types

actually do fgrep. -G -F and -E are now mutally exclusive, and override
the program name as expected. ok millert@

strncpy -> memcpy per deraadt suggestion.
also add a note why we can't use strlcpy.

go back to using strncpy. for long patterns, strlcpy reads too much
of a potentially nontermined src.

faster grep for simple patterns. derived from a patch by sean farley.
this makes searching for constant strings much faster by avoiding regex.
ok deraadt@

minor tweaks

use strlcpy, not strncpy. ok deraadt@ millert@