.\" awkrsall(1) — comprehensive all-in-one man page for awkrs
.\" Modeled after zshall(1): every man page in one document.
.\" Render with: man -l man/man1/awkrsall.1
.\" Install: cp man/man1/awkrsall.1 /usr/local/share/man/man1/
.TH AWKRSALL 1 "2026-05-31" "awkrs 0.4.13" "User Commands"
.SH NAME
awkrsall \- the awkrs comprehensive reference (invocation, language,
bytecode VM, Cranelift JIT, persistent bytecode cache, parallel records,
gawk/mawk parity)
.SH OVERVIEW
This is the all-in-one
.B awkrs
manual, modeled on
.BR zshall (1).
It concatenates the major sections of the awkrs reference into a single
page so you can grep with `/`:
.RS 2
.PP
.B 1. INVOCATION
.RB "\(em " "POSIX, GNU, gawk, mawk -W flags"
.br
.B 2. PROGRAM TEXT
.RB "\(em " "rules, patterns, statements, getline, operators"
.br
.B 3. DATA MODEL
.RB "\(em " "fields, arrays, records, FS/FPAT/RS, special vars"
.br
.B 4. BUILTINS
.RB "\(em " "every string/numeric/I/O/bit builtin"
.br
.B 5. GAWK EXTENSIONS
.RB "\(em " "@include, @load, @namespace, /inet/, indirect calls"
.br
.B 6. BYTECODE VM
.RB "\(em " "stack VM, opcodes, peephole fusion, inline fast paths"
.br
.B 7. CRANELIFT JIT
.RB "\(em " "tiered compilation, ABI, eligible ops, NaN-boxing"
.br
.B 8. BYTECODE CACHE
.RB "\(em " "rkyv, mmap, atomic rename, mtime invalidation"
.br
.B 9. PARALLEL RECORDS
.RB "\(em " "static safety check, rayon, deterministic reorder"
.br
.B 10. BENCHMARKS
.RB "\(em " "vs gawk / mawk / BSD awk / frawk / goawk"
.br
.B 11. ENVIRONMENT
.br
.B 12. EXIT STATUS
.br
.B 13. FILES
.br
.B 14. EXAMPLES
.RB "\(em " "throughput, CSV, parallel, JIT-friendly, regex"
.br
.B 15. PARITY GAPS
.RE
.PP
For a shorter quick-reference page see
.BR awkrs (1).
.SH NAME
awkrs \- pattern-directed scanning and processing language with
bytecode VM, Cranelift JIT, persistent rkyv bytecode cache, and parallel
record processing (Rust awk implementation)
.SH SYNOPSIS
.B awkrs
.RI [ OPTIONS ]
.RB [ \-f
.IR PROGFILE
.RB |
.B \-e
.IR PROGRAM
.RB |
.IR PROGRAM ]
.RB [ \-\- ]
.RI [ file ...]
.SH DESCRIPTION
.B awkrs
runs
.B pattern \(-> action
programs over input records like POSIX
.BR awk ,
GNU
.BR gawk ,
and
.BR mawk .
The CLI accepts the union of POSIX, gawk, and mawk
.RB ( "\-W" )
options.
.PP
Execution flows through a Cranelift-JIT bytecode VM with a parallel
record processor that activates automatically when the program is
parallel-safe and
.B \-j
selects more than one thread; otherwise execution is sequential.
.PP
Source survey claim: among the major public awk implementations
(BWK awk, gawk, mawk, goawk, frawk, zawk),
.B awkrs is the first to combine a bytecode VM, a JIT compiler, and a
.B persistent on-disk bytecode cache.
frawk has VM + Cranelift/LLVM JIT but recompiles every invocation;
gawk's
.B pm-gawk
persists script-defined variables, not compiled bytecode.
.SH "1. INVOCATION"
The program text is taken from the first non-option argument unless
.B \-f
or
.B \-e
is given. Remaining non-option arguments are input files; if none are
given, standard input is read.
.SS POSIX
.TP
.BR \-f " \fIPROGFILE\fR, " \-\-file =\fIPROGFILE\fR
Read program source from
.IR PROGFILE .
Repeatable; sources concatenated.
.TP
.BR \-F " \fIFS\fR, " \-\-field-separator =\fIFS\fR
Set input field separator
.RB ( FS ).
.TP
.BR \-v " \fIvar=val\fR, " \-\-assign =\fIvar=val\fR
Assign
.I val
to
.I var
before
.B BEGIN
runs. Repeatable.
.SS GNU program sources
.TP
.BR \-e " \fIPROGRAM\fR, " \-\-source =\fIPROGRAM\fR
Inline program text. Repeatable; sources concatenated.
.TP
.BR \-i " \fIFILE\fR, " \-\-include =\fIFILE\fR
Include awk source file (gawk
.B @include
behavior). Repeatable.
.SS gawk extensions
.TP
.BR \-b ", " \-\-characters-as-bytes
Use byte length for
.BR length ,
.BR substr ,
.BR index .
.TP
.BR \-c ", " \-\-traditional
Traditional POSIX awk mode.
.TP
.BR \-C ", " \-\-copyright
Print copyright information.
.TP
.BR \-d "[\fIFILE\fR], " \-\-dump-variables [=\fIFILE\fR]
Dump globals after the program runs.
.TP
.BR \-D "[\fIFILE\fR], " \-\-debug [=\fIFILE\fR]
Static rule/function listing. Not gawk's interactive debugger.
.TP
.BR \-E " \fIFILE\fR, " \-\-exec =\fIFILE\fR
Execute program from
.I FILE
then exit.
.TP
.BR \-g ", " \-\-gen-pot
Generate gettext
.B .pot
output and exit.
.TP
.BR \-I ", " \-\-trace
Trace execution.
.TP
.BR \-k ", " \-\-csv
CSV mode. Sets
.B FS / FPAT
and uses a parser aligned with
.BR "gawk --csv" .
.TP
.BR \-l " \fILIB\fR, " \-\-load =\fILIB\fR
Load extension library (gawk
.BR @load
behavior). Repeatable.
.TP
.BR \-L "[\fIVAL\fR], " \-\-lint [=\fIVAL\fR]
Lint checks.
.TP
.BR \-M ", " \-\-bignum
MPFR arbitrary-precision arithmetic.
.TP
.BR \-n ", " \-\-non-decimal-data
Accept hex (\fB0x\fP) and octal (leading \fB0\fP) input numbers.
.TP
.BR \-N ", " \-\-use-lc-numeric
Respect
.B LC_NUMERIC
for
.BR sprintf / printf .
.TP
.BR \-o "[\fIFILE\fR], " \-\-pretty-print [=\fIFILE\fR]
Pretty-print parsed program text.
.TP
.BR \-O ", " \-\-optimize
Bytecode optimizations on (default).
.TP
.BR \-p "[\fIFILE\fR], " \-\-profile [=\fIFILE\fR]
Profile execution.
.TP
.BR \-P ", " \-\-posix
Strict POSIX mode.
.TP
.BR \-r ", " \-\-re-interval
Enable interval expressions in regex.
.TP
.BR \-s ", " \-\-no-optimize
Disable bytecode optimizations.
.TP
.BR \-S ", " \-\-sandbox
Disable
.BR system ,
.BR getline " | cmd" ,
and other shell-out features.
.TP
.BR \-t ", " \-\-lint-old
Lint against historical awk dialects.
.SS mawk
.TP
.BR \-W " \fIspec\fR"
Mawk-style options forwarded to the runtime
.RB ( "compat" ,
.BR "interactive" ,
.BR "sprintf=N" ,
.BR "exec=FILE" ,
.BR "posix_space" ,
.BR "dump" ,
etc.).
.SS awkrs-specific
.TP
.BR \-j " \fIN\fR, " \-\-threads =\fIN\fR
Worker thread count. Default 1. Activates parallel record processing
when the program is parallel-safe (see
.B "9. PARALLEL RECORDS" ).
.TP
.BR \-\-read-ahead " \fILINES\fR"
Stdin parallel batch size. Default 1024.
.TP
.B \-\-no-jit
Force the bytecode interpreter even when JIT is available.
.TP
.B \-\-no-cache
Disable bytecode cache for this invocation.
.SH "2. PROGRAM TEXT"
.SS Rules
.BR BEGIN ,
.BR END ,
.BR BEGINFILE ,
.BR ENDFILE ,
empty pattern,
.BR /regex/ ,
expression patterns, range patterns
.RB ( /a/,/b/
or
.BR "NR==1,NR==5" ).
The four special patterns
.B must
use
.BR { \ ...\ } ;
record rules may omit braces for the default
.BR "{ print $0 }" .
.SS Statements
.BR if ,
.BR while ,
.B do\-while ,
.B for
(C-style and
.BR "for (i in arr)" ),
.BR switch / case / default
(gawk-style: no fall-through, regex
.BR "case /re/" ),
.BR print / printf
with
.BR > ,
.BR >> ,
.BR | ,
.B |&
redirection,
.BR break ,
.BR continue ,
.BR next ,
.BR nextfile ,
.BR exit ,
.BR delete ,
.BR return ,
.BR getline .
.SS getline as expression
Returns
.B 1
(read),
.B 0
(EOF),
.B \-1
(error),
.B \-2
(gawk retryable I/O when
.BR PROCINFO[input,"RETRY"]
is set). JIT
.B getline
failures abort the JIT chunk instead of returning
.BR \-1 / \-2 .
.SS Operators
arithmetic, comparison, string concat, ternary,
.BR in ,
.BR ~ / !~ ,
.BR ++ / \-\-
(prefix/postfix on vars,
.BR $n ,
.BR a[k] ),
.BR ^ / **
(right-associative; unary
.BR + / \- / !
bind looser, so
.B \-2^2
=
.BR \-(2^2) ).
.B Division by zero
.RB ( / ,
compound
.BR /= )
is a
.B fatal error
.RB ( "division by zero attempted" ),
not infinity.
.SS Primary /
The lexer may emit
.B /
as division when
.B regex_mode
is false (e.g. after
.BR = ).
At a
.B primary
position
.B /
cannot start division, so the parser re-reads as
.BR /regex/ .
In
.B expression
context a bare
.B /re/
means
.B $0 ~ /re/
(POSIX), so
.BR "!/foo/" ,
.BR "if (/foo/)" ,
.BR "x = /foo/"
use a match against the current record.
.SS gawk regexp constants
.B @/pattern/
yields a
.B regexp
value
.RB ( typeof
reports
.BR regexp );
.B ~
uses the pattern as a regex.
.SH "3. DATA MODEL"
Fields, scalars, associative arrays
.RB ( a[k] ,
.B a[i,j]
with
.BR SUBSEP ),
.BR ARGC / ARGV
(set before
.B BEGIN
\(em
.B ARGV[0]
is the executable,
.B ARGV[1..]
are file paths).
.PP
.BR FS
is a regex when multi-char.
.B FPAT
(gawk-style): non-empty splits by regex match.
.BR split / patsplit
accept regex as 3rd arg;
.B patsplit
4-arg form populates
.BR seps .
.PP
.B POSIX record model:
.B NF = n
truncates or extends fields and rebuilds
.B $0
with
.BR OFS ;
.B "$0 = ..."
re-splits and updates
.BR NF .
.PP
.B FS/FPAT from literals:
bytecode may store source
.B "..."
as an internal literal string, but
.B cached_fs
sync on read still tracks those assignments.
.PP
.B Whole array in a scalar context
.RB ( "print a" ,
string concat,
.B printf
args,
.B ~
operands, etc.) is a
.B fatal runtime error
(gawk-style), not a silent empty string.
.PP
.B Scalars:
uninitialized variables compare like numeric
.B 0
where POSIX expects dual
.B 0 / ""
.
.PP
.B String constants vs input:
program string literals are not
.I numeric strings
for
.BR < / <= / > / >=
the way
.B $n
can be; arithmetic still uses longest-prefix string\(->number
.RB ( "\"3.14abc\"+0"
=
.BR 3.14 ).
.PP
.B split("", arr, fs)
returns
.B 0
(no empty pseudo-field).
.SS Records & env
.B RS
is newline by default; one (UTF-8) char is a literal delimiter;
.B RS=""
is paragraph mode; multi-char is a gawk regex
.RB ( RT
holds the matched text).
.B FIELDWIDTHS
selects fixed-width when non-empty.
.PP
.BR ENVIRON ,
.BR CONVFMT ,
.BR OFMT ,
.BR FIELDWIDTHS ,
.B IGNORECASE
(case-insensitive regex +
.BR == / !=
ordering via
.BR strcoll ),
.BR ARGIND ,
.BR ERRNO ,
.BR LINT ,
.BR TEXTDOMAIN ,
.BR BINMODE .
.BR PROCINFO ,
.BR FUNCTAB ,
.B SYMTAB
populated in gawk-compatible form.
.SH "4. BUILTINS"
.B length ,
.B index
(empty needle =
.BR 1 ,
matching gawk),
.B substr
.RB ( "gawk rule" ,
not POSIX: if
.B "start < 1"
clamp to
.B 1
and leave
.B length
unchanged),
.BR intdiv ,
.BR mkbool ,
.BR split ,
.BR sprintf ,
.BR printf
(flags,
.B *
and
.B %n$
positional, gawk
.BR %' ,
conversions
.BR "%s %d %i %u %o %x %X %f %e %E %g %G %c %%"
\(em
.BR %e / %E
use signed two-digit exponents;
.B %c
uses a string's first character),
.BR gsub ,
.BR sub ,
.BR match ,
.BR gensub ,
.BR isarray ,
.BR tolower ,
.BR toupper ,
.BR int .
.PP
Math:
.BR sin ,
.BR cos ,
.BR atan2 ,
.BR exp ,
.BR log ,
.BR sqrt .
.PP
.BR rand ,
.BR srand ,
.BR systime ,
.BR strftime
(0\(en3 args),
.BR mktime ,
.BR system ,
.BR close ,
.BR fflush ,
bit ops
.RB ( and ,
.BR or ,
.BR xor ,
.BR lshift ,
.BR rshift ,
.BR compl ),
.BR strtonum ,
.BR asort ,
.BR asorti .
.PP
User-defined
.B function
with parameter locals. Recursion is capped (256 nested calls in
release builds; lower in unit tests) so pathological self-calls fail
with an error instead of overflowing the host stack.
.SH "5. GAWK EXTENSIONS"
.BR @include ,
.B @load "*.awk" ,
.B @namespace "..."
(default identifier prefixing; built-ins exempt), indirect calls
.RB ( @name(...) ,
.BR @(expr)(...) ),
.B /inet/tcp/...
and
.B /inet/udp/...
client sockets, gettext builtins
.RB ( bindtextdomain ,
.BR dcgettext ,
.B dcngettext
with
.B .mo
catalogs via the
.B gettext
crate),
.BR \-M / \-\-bignum
MPFR.
.SH "6. BYTECODE VM"
.B awkrs
compiles AWK programs into a flat bytecode instruction stream and runs
them on a
.B stack VM.
Short-circuit
.BR && / || ,
control flow, and range patterns resolve to jump-patched offsets at
compile time. The
.B string pool
interns variable names and string constants for cheap
.B u32
indexing.
.SS Peephole fusion
Common sequences combine into single opcodes:
.PP
.TS
tab(|);
l l.
.B Source pattern|.B Fused opcode
\fBprint $N\fR|\fBPrintFieldStdout\fR (zero-alloc field write)
\fBs += $N\fR|\fBAddFieldToSlot\fR (in-place numeric parse)
\fBi = i + 1\fR / \fBi++\fR / \fB++i\fR|\fBIncrSlot\fR (one numeric add, no stack traffic)
\fBs += i\fR between slots|\fBAddSlotToSlot\fR
\fB$1 "," $2\fR literal concat|\fBConcatPoolStr\fR
\fBNR++\fR HashMap-path|\fBIncrVar\fR
.TE
.SS Inline fast paths
Single-rule programs with one fused opcode
.RB ( "{ print $1 }" ,
.BR "{ s += $1 }" )
bypass
.B VmCtx
entirely.
.PP
Memory-mapped files also recognize
.B "{ gsub(\"lit\", \"repl\"); print }"
with literal pattern: when the needle is absent, the loop writes each
line from the mapped buffer with
.B ORS
and skips the VM.
.SH "7. CRANELIFT JIT"
The
.B jit
module compiles eligible bytecode chunks to native code using Cranelift
with
.BR "opt_level=speed" .
.SS ABI
.B "(vmctx, slots, field_fn, array_field_add, var_dispatch, field_dispatch, io_dispatch, val_dispatch) -> f64"
\(em opaque
.B vmctx
plus seven
.B "extern \"C\""
callbacks (no thread-local lookups).
.SS Tiered compilation
Chunks count entries and only compile after
.B AWKRS_JIT_MIN_INVOCATIONS
(default 3; set
.B 1
for first-entry compile). Set
.B AWKRS_JIT=0
to force the bytecode interpreter.
.SS Eligible JIT ops
Constants, arithmetic/comparisons, jumps and fused loop tests, slot and
HashMap-path scalar ops, field reads (including dynamic
.B $i
writes in mixed mode),
.BR for-in ,
.BR asort ,
.BR asorti ,
.BR match ,
.BR sub ,
.BR gsub ,
.BR split ,
.BR patsplit ,
.B getline
(primary, file, coproc),
.BR CallUser ,
fused print opcodes (including redirects),
.BR typeof ,
and a whitelist of builtins (math,
.BR sprintf ,
.BR printf ,
.BR strftime ,
.BR fflush ,
.BR close ,
.BR system ).
.SS Value representation
Mixed-mode chunks NaN-box
.B Value
in slots; non-mixed chunks keep slots in Cranelift SSA. Unsupported
opcodes fall back to the bytecode loop.
.SH "8. BYTECODE CACHE"
.B "\-f script.awk"
invocations memoize the compiled
.B CompiledProgram
to
.B ~/.awkrs/scripts.rkyv
\(em an rkyv-archived shard with
.B mmap
+ zero-copy
.B ArchivedHashMap
lookup on the read path
.RB ( "check_archived_root"
validation) and
.BR flock -serialized
atomic-rename writes.
.PP
Each entry's inner
.B CompiledProgram
blob is bincode (rkyv outer, bincode inner \(em same architecture as
.BR zshrs
and
.BR strykelang ). Repeat runs skip lex/parse/compile entirely \(em
only the matched entry's blob is decoded.
.PP
.B Invalidation:
entries are invalidated on source-file mtime change or when the running
.B awkrs
binary is newer than the cached entry. Any rebuild silently rebuilds
the cache.
.PP
.B Disable
with
.BR "AWKRS_CACHE=0" .
The cache only engages for the simple
.B "\-f script.awk"
form \(em inline
.BR \-e / \-\-source ,
.BR \-E ,
.BR \-i / \-\-include ,
.BR \-l / \-\-load ,
.BR \-\-debug ,
.BR \-\-lint ,
.BR \-\-pretty-print ,
.B \-\-gen-pot
skip the cache because they need the AST.
.SH "9. PARALLEL RECORDS"
.PP
.RS 2
.nf
┌─────────────────────────────────────────────┐
│ WORKER 0 ░░ CHUNK 0 ██ REORDER QUEUE │
│ WORKER 1 ░░ CHUNK 1 ██ ──────────────>│
│ WORKER 2 ░░ CHUNK 2 ██ DETERMINISTIC │
│ WORKER N ░░ CHUNK N ██ OUTPUT STREAM │
└─────────────────────────────────────────────┘
.fi
.RE
.PP
Default
.B "\-j / \-\-threads"
is
.BR 1 .
Pass a higher value when the program is
.B parallel-safe
\(em static check:
.RS 2
.PP
no range patterns
.br
no
.BR exit ,
.BR nextfile ,
.B delete
.br
no primary
.B getline
.br
no pipe/coproc
.B getline
.br
no
.BR asort / asorti
.br
no indirect calls
.br
no
.BR print / printf
redirection
.br
no cross-record assignments
.RE
.PP
Records are processed in parallel via
.B rayon
and output is
.B reordered to input order
within each batch so pipelines stay deterministic.
.SS File input
.B Regular files
are memory-mapped
.RB ( memmap2 )
and scanned with the same
.B RS
rules as the sequential path \(em no
.BR read()
copy of the whole file.
.SS Stdin
.B Stdin parallel
chunks up to
.B \-\-read-ahead
lines (default 1024) per batch, dispatches to workers, emits in order,
then refills.
.SS Worker model
Workers run the same bytecode VM as the sequential path. The compiled
program is shared via
.B "Arc<CompiledProgram>"
(one compile, cheap refcount per worker) with per-worker runtime state.
.SS Fallback
Non-parallel-safe programs run sequentially with a warning when
.BR "\-j > 1" .
Programs that use primary
.B getline
(including in
.BR BEGIN )
also run sequentially for file input.
.B END
only sees post-
.B BEGIN
global state \(em record-rule mutations from parallel workers are not
merged.
.SH "10. BENCHMARKS"
Combat metrics on representative workloads (1 million lines unless
noted).
.B awkrs
ties or beats every other awk implementation surveyed on most
workloads. See README
.B "[0x06] BENCHMARKS"
for the full table with timings. Highlights:
.PP
.TS
tab(|);
l l.
.B Benchmark|.B Winner
\fB{ print $1 }\fR throughput|awkrs (fused PrintFieldStdout)
CPU-bound \fBBEGIN\fR (no input)|awkrs / frawk near-tie
Sum first column|awkrs (AddFieldToSlot)
Regex filter (no matches)|awkrs / mawk
Associative array build|awkrs / gawk
\fBgsub\fR no-match (mmap fast path)|awkrs (mapped-buffer write, VM skipped)
.TE
.SH "11. ENVIRONMENT"
.TP
.B AWKPATH
Search path for
.B \-l
and
.BR @include .
Default
.BR . .
.TP
.B AWKLIBPATH
Search path for
.BR @load .
.TP
.B AWKRS_CACHE
Set to
.B 0
to disable the bytecode cache.
.TP
.B AWKRS_JIT
Set to
.B 0
to force the bytecode interpreter.
.TP
.B AWKRS_JIT_MIN_INVOCATIONS
JIT tier-up threshold (default
.BR 3 ).
.TP
.B GAWK_READ_TIMEOUT
Default read timeout (ms) used as the fallback for
.BR PROCINFO[input,"READ_TIMEOUT"] .
.TP
.B LC_NUMERIC
Locale decimal radix and thousands grouping when
.B \-N
is set; affects
.BR sprintf / printf / print
output only, not
.BR $n / $0
parsing.
.SH "12. EXIT STATUS"
.TP
.B 0
Successful completion.
.TP
.B 1
Error during program execution or invalid usage. Errors are written to
stderr as
.BR "awkrs: <command>: <reason>" .
.TP
.I n
The program may exit with any status by calling
.BR "exit n" .
.SH "13. FILES"
.TP
.I ~/.awkrs/scripts.rkyv
Persistent bytecode cache shard. rkyv outer + bincode inner;
mmap+validate on read,
.BR flock -serialized
atomic rename on write.
.TP
.I /usr/local/share/man/man1/awkrs.1
Short reference man page.
.TP
.I /usr/local/share/man/man1/awkrsall.1
This man page.
.SH "14. EXAMPLES"
.PP
Print the second whitespace-separated field:
.RS
.nf
awkrs '{ print $2 }' input.txt
.fi
.RE
.PP
Sum a CSV column with quoted-field handling and four worker threads:
.RS
.nf
awkrs -k -j4 '{ s += $3 } END { print s }' data.csv
.fi
.RE
.PP
Memoize compiled bytecode on disk:
.RS
.nf
awkrs -f myscript.awk input.txt
# second invocation skips lex / parse / compile entirely
.fi
.RE
.PP
Force the bytecode interpreter (no JIT):
.RS
.nf
AWKRS_JIT=0 awkrs '{ print $2 }' input.txt
.fi
.RE
.PP
First-entry JIT (tier-up threshold = 1):
.RS
.nf
AWKRS_JIT_MIN_INVOCATIONS=1 awkrs -f hot.awk huge.log
.fi
.RE
.PP
CSV mode with named fields:
.RS
.nf
awkrs -k '{ print $1, $5 }' data.csv
.fi
.RE
.PP
TCP client socket:
.RS
.nf
awkrs 'BEGIN { "/inet/tcp/0/example.com/80" |& getline line; print line }'
.fi
.RE
.PP
Indirect function call:
.RS
.nf
awkrs 'function f(x) { return x*2 } BEGIN { name = "f"; print @name(21) }'
.fi
.RE
.PP
gawk regexp constant:
.RS
.nf
awkrs 'BEGIN { r = @/^[0-9]+$/; print typeof(r) }'
.fi
.RE
.PP
Parallel-safe sum across 8 threads:
.RS
.nf
awkrs -j8 '{ s += $1 } END { print s }' big.log
.fi
.RE
.PP
.B BEGINFILE
hook:
.RS
.nf
awkrs 'BEGINFILE { print "opening", FILENAME } /pattern/ { print }' *.log
.fi
.RE
.PP
.B \-\-debug
static listing:
.RS
.nf
awkrs --debug=program.lst -f myscript.awk
.fi
.RE
.PP
MPFR arbitrary precision:
.RS
.nf
awkrs -M 'BEGIN { PREC = 200; print 1 / 7 }'
.fi
.RE
.PP
Locale-aware printf:
.RS
.nf
LC_NUMERIC=fr_FR.UTF-8 awkrs -N 'BEGIN { printf("%.2f\\n", 1234.5) }'
.fi
.RE
.SH "15. PARITY GAPS"
.B awkrs
implements the union of POSIX, gawk, and mawk. Remaining gaps are
documented in the project compatibility matrix
.RB ( docs/COMPATIBILITY.md ).
Notable items:
.IP \[bu] 2
.B \-D
.BR \-\-debug
emits a static rule/function listing; it is
.B not
gawk's interactive debugger (no breakpoints, no step / next).
.IP \[bu] 2
.B \-c
.BR \-\-traditional
is stored on the runtime; effect today is minimal.
.IP \[bu] 2
Some
.B PROCINFO[input,...]
keys (e.g.
.BR RETRY ,
.BR READ_TIMEOUT )
are honored only on retryable I/O paths.
.IP \[bu] 2
Cross-record state mutations in parallel mode are not merged into
.BR END ;
the static safety check refuses such programs unless
.B "\-j 1"
is used.
.SH SEE ALSO
.BR awkrs (1),
.BR awk (1),
.BR gawk (1),
.BR mawk (1),
.BR strykeall (1),
.BR zshrsall (1),
.BR lsofrsall (1),
.BR temprsall (1)
.SH AUTHOR
.B Jacob Menke
.RI < https://github.com/MenkeTechnologies >
.SH BUGS
Report bugs at
.UR https://github.com/MenkeTechnologies/awkrs/issues
.UE .