Make cronic portable to macOS's BSD grep and cmp, and test it on macOS

.github/workflows/tests.yaml

@@ -34,3 +34,19 @@ jobs:
           command -v xxd
           command -v bc
       - run: make -C tests test
+  # This job tests `cronic` under macOS's /bin/bash, which is bash 3.2, and
+  # macOS's BSD utilities.  `cronic` and the bash scripts in its tests run
+  # under the first `bash` in PATH, so /bin is put first, ahead of any
+  # Homebrew bash.
+  cronic-macos:
+    runs-on: macos-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          fetch-depth: 1
+          show-progress: false
+          persist-credentials: false
+      - run: echo /bin >> "$GITHUB_PATH"
+      - run: bash --version
+      - run: make -C tests/cronic-test test

cronic

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 
 # Cronic v2 - cron job report wrapper
 # Copyright 2007 Chuck Houpt
@@ -14,8 +14,10 @@
 # --duplicate-stderr options, which may be given in any order.
 # The options and the treatment of execution traces are documented in function
 # `cronic_help`; run `cronic --help`.
-# This script runs under bash 3.2, which is /bin/bash on macOS, so it uses a
-# feature of a later bash only after checking the version.
+# This script runs under the first `bash` in PATH, which on macOS may be
+# /bin/bash, bash 3.2.  So it uses a feature of a later bash, such as
+# BASH_XTRACEFD (bash 4.1), only after checking the version; see the
+# BASH_VERSINFO check below.
 
 # Exit if a command exits with a non-zero status, and treat unset variables
 # as an error.
@@ -174,6 +176,15 @@ set -e
 # Unsetting BASH_XTRACEFD also closes its file descriptor.
 unset BASH_XTRACEFD
 
+# End stderr with a newline, if it is not empty.  `grep` ends its output with a
+# newline, so without this, the filtered stderr would differ from the
+# unfiltered stderr even when `grep` removed no line, and the report would
+# show the same output twice.  The newline also keeps the last line of stderr
+# off the line of the report's next heading.
+if [ -s "$cronic_trace" ] && [ "$(tail -c 1 "$cronic_trace" | wc -l)" -eq 0 ]; then
+  echo >> "$cronic_trace"
+fi
+
 # Prints a basic regular expression that matches its argument literally.
 # The argument must not contain a backslash.
 cronic_bre_quote() {
@@ -198,16 +209,26 @@ cronic_trace_pattern() {
   local LC_ALL=C
   [[ $PS4 =~ ^([$'\xc0'-$'\xff'][$'\x80'-$'\xbf']*|.) ]]
   local first=${BASH_REMATCH[0]}
-  printf '^\\(%s\\)\\+%s\n' "$(cronic_bre_quote "$first")" "$(cronic_bre_quote "${PS4#"$first"}")"
+  local rest=${PS4#"$first"}
+  # `\{1,\}` rather than GNU's `\+`, which is not a POSIX basic regular
+  # expression.
+  printf '^\\(%s\\)\\{1,\\}%s\n' "$(cronic_bre_quote "$first")" "$(cronic_bre_quote "$rest")"
 }
 
+# Each `grep` that reads the wrapped command's output uses `-a`, so that it
+# treats the output as text even if it contains a NUL byte or a byte that is
+# invalid in the locale.  Without `-a`, `grep` would treat such output as
+# binary and print only a "Binary file ... matches" notice in place of the
+# lines it selects, so that a real error could go unreported.
 cronic_pattern=$(cronic_trace_pattern)
-if [ -n "$cronic_pattern" ] && grep -q "$cronic_pattern" "$cronic_trace"; then
-  # `|| true` because `grep -v` exits with status 1 when it selects no lines,
-  # which happens whenever stderr consists only of trace lines.  Without it,
-  # `set -e` would abort the script here, before it produces its report, and
-  # it would exit with status 1 rather than with the wrapped command's status.
-  grep -v "$cronic_pattern" "$cronic_trace" > "$cronic_err" || true
+if [ -n "$cronic_pattern" ]; then
+  # `|| [ $? -eq 1 ]` because `grep -v` exits with status 1 when it selects no
+  # lines, which happens whenever stderr consists only of trace lines.  Without
+  # it, `set -e` would abort the script here, before it produces its report,
+  # and it would exit with status 1 rather than with the wrapped command's
+  # status.  If `grep` fails (status 2), the unfiltered stderr is used, rather
+  # than an empty or truncated file cronic_err that could hide an error.
+  grep -a -v "$cronic_pattern" "$cronic_trace" > "$cronic_err" || [ $? -eq 1 ] || cronic_err=$cronic_trace
 else
   cronic_err=$cronic_trace
 fi
@@ -216,11 +237,17 @@ fi
 # For example, git does this, and the POSIX standard requires it for make.
 # The full stderr output is shown, but file cronic_reduced_err is checked for
 # emptiness below.
+# GNU make names itself by the name it was invoked as, without the directory,
+# such as `make`, `gmake`, `gnumake` (on macOS), or `mingw32-make`.  It quotes
+# the directory as 'dir' from version 4.0, and as `dir' before then; macOS's
+# /usr/bin/make is GNU make 3.81.
 # TODO: Expand this.
-cat "$cronic_err" \
-  | grep -v "^Everything up-to-date$" \
-  | grep -v '^make\(\[[0-9]\+\]\)\?: \(Entering\|Leaving\) directory '"'"'.*'"'"'$' \
-    > "$cronic_reduced_err" || true
+# `|| [ $? -eq 1 ]` because `grep -v` exits with status 1 when it selects no
+# lines.  If `grep` fails (status 2), the unreduced stderr is used.
+grep -a -E -v \
+  -e '^Everything up-to-date$' \
+  -e '^[[:alnum:]_.-]*make(\[[0-9]+\])?: (Entering|Leaving) directory '"[\`']"'.*'"'"'$' \
+  "$cronic_err" > "$cronic_reduced_err" || [ $? -eq 1 ] || cronic_reduced_err=$cronic_err
 
 if [ $cronic_result -ne "$cronic_expected" ] || { [ $cronic_permitstderr -eq 0 ] && [ -s "$cronic_reduced_err" ]; }; then
   echo "Cronic detected failure or error output for the command:"
@@ -240,15 +267,15 @@ if [ $cronic_result -ne "$cronic_expected" ] || { [ $cronic_permitstderr -eq 0 ]
   cat "$cronic_err"
   echo
   if [ "$cronic_debug" != "true" ]; then
-    if ! cmp --silent "$cronic_err" "$cronic_reduced_err"; then
+    if ! cmp -s "$cronic_err" "$cronic_reduced_err"; then
       echo "REDUCED ERROR OUTPUT:"
       cat "$cronic_reduced_err"
       echo
     fi
   fi
   echo "STANDARD OUTPUT:"
   cat "$cronic_out"
-  if [ "$cronic_trace" != "$cronic_err" ]; then
+  if ! cmp -s "$cronic_trace" "$cronic_err"; then
     echo
     echo "TRACE-ERROR OUTPUT:"
     cat "$cronic_trace"

tests/cronic-test/cronic-test.sh

@@ -10,6 +10,13 @@
 # printed, and the exit status would be grep's 1 rather than the wrapped
 # command's.  (The temporary files would still be removed, by `cronic`'s EXIT
 # trap, so the left-behind-files check in `check` does not detect that bug.)
+# The checks that catch that bug are the "trace-only stderr" checks that use
+# `stderr-lines` with a literal PS4, and the "nested trace in stderr" check,
+# whose bash script writes its trace lines to stderr whatever the bash version.
+# All of them run only when the test is not run as root.  The checks that use
+# `trace-only`, a bash script, catch it only under a bash earlier than 4.1:  a
+# later bash writes its trace lines to BASH_XTRACEFD, so `cronic` sees no trace
+# lines in stderr and does not run `grep -v`.
 
 set -eu
 
@@ -30,7 +37,7 @@ status=0
 
 # A command whose stderr is nothing but trace lines.
 cat > "$work/trace-only" << 'EOF'
-#!/bin/bash
+#!/usr/bin/env bash
 set -x
 echo "the standard output"
 exit "$1"
@@ -39,20 +46,30 @@ chmod +x "$work/trace-only"
 
 # A command that writes real error output in addition to trace lines.
 cat > "$work/trace-and-stderr" << 'EOF'
-#!/bin/bash
+#!/usr/bin/env bash
 set -x
 echo "a real error" 1>&2
 exit "$1"
 EOF
 chmod +x "$work/trace-and-stderr"
 
 # A command whose stderr is nothing but `make` directory-change notices.  Both
-# the top-level form (`make:`) and the recursive form (`make[1]:`) appear; both
-# must be filtered out of the reduced error output.
+# the top-level form (`make:`) and the recursive form (`make[1]:`) appear, as do
+# the quoting of GNU make 3.81 (macOS's /usr/bin/make) and the other names
+# that GNU make runs under; all must be filtered out of the reduced error
+# output.
 cat > "$work/make-noise" << 'EOF'
 #!/bin/sh
 echo "make: Entering directory '/tmp/x'" 1>&2
 echo "make[1]: Entering directory '/tmp/x/sub'" 1>&2
+echo 'make[2]: Entering directory `/tmp/x/sub/old'"'" 1>&2
+echo 'make[2]: Leaving directory `/tmp/x/sub/old'"'" 1>&2
+echo "gmake[2]: Entering directory '/tmp/x/sub/g'" 1>&2
+echo "gmake[2]: Leaving directory '/tmp/x/sub/g'" 1>&2
+echo "gnumake[2]: Entering directory '/tmp/x/sub/gnu'" 1>&2
+echo "gnumake[2]: Leaving directory '/tmp/x/sub/gnu'" 1>&2
+echo "mingw32-make[2]: Entering directory '/tmp/x/sub/mingw'" 1>&2
+echo "mingw32-make[2]: Leaving directory '/tmp/x/sub/mingw'" 1>&2
 echo "make[1]: Leaving directory '/tmp/x/sub'" 1>&2
 echo "make: Leaving directory '/tmp/x'" 1>&2
 exit "$1"
@@ -71,6 +88,28 @@ exit "$1"
 EOF
 chmod +x "$work/make-noise-and-stderr"
 
+# A command that writes a trace line and then an error line that contains a
+# NUL byte and a byte that is invalid in UTF-8, so that `grep` would treat the
+# output as binary.
+cat > "$work/binary-stderr" << 'EOF'
+#!/bin/sh
+printf '[trace] echo hi\n' 1>&2
+printf 'bad \377\000 error\n' 1>&2
+EOF
+chmod +x "$work/binary-stderr"
+
+# A bash command whose trace lines go to stderr, as under a bash earlier than
+# 4.1, whatever the bash version.  Its trace lines are nested 3 deep.
+cat > "$work/nested-trace" << 'EOF'
+#!/usr/bin/env bash
+unset BASH_XTRACEFD
+set -x
+outer=$(inner=$(echo deepest); echo "$inner")
+echo "$outer"
+exit "$1"
+EOF
+chmod +x "$work/nested-trace"
+
 # A command that writes, to stderr, lines that start with `+` but are not trace
 # lines.
 cat > "$work/plus-stderr" << 'EOF'
@@ -96,9 +135,10 @@ temp_files() {
 # check DESCRIPTION EXPECTED-STATUS EXPECTED-OUTPUT COMMAND...: runs `cronic`
 # on COMMAND and checks its exit status, whether it printed a report, and that
 # it left no temporary files behind.  EXPECTED-OUTPUT is "silent", "report",
-# "report:TEXT", which also requires TEXT to appear in the report, or
-# "message:TEXT", which requires TEXT to appear in output that need not be a
-# report.
+# "report:TEXT", which also requires TEXT to appear in the report,
+# "report-once:TEXT", which also requires TEXT to appear on exactly one line of
+# the report, or "message:TEXT", which requires TEXT to appear in output that
+# need not be a report.
 check() {
   description="$1"
   expected_status="$2"
@@ -136,6 +176,15 @@ check() {
       ;;
   esac
   case "$expected_output" in
+    report-once:*)
+      if [ "$(grep -c -e "${expected_output#*:}" "$work/output")" -ne 1 ]; then
+        echo "FAIL: $description: expected exactly one line of the output to" \
+          "contain \"${expected_output#*:}\", but got:"
+        cat "$work/output"
+        status=1
+        return
+      fi
+      ;;
     report:* | message:*)
       if ! grep -q -e "${expected_output#*:}" "$work/output"; then
         echo "FAIL: $description: expected the output to contain" \
@@ -183,6 +232,20 @@ check "make directory-change notices, exit 0" 0 silent "$work/make-noise" 0
 check "make directory-change notices and real stderr, exit 0" 0 \
   "report:^a real error$" "$work/make-noise-and-stderr" 0
 
+# Error output that does not end with a newline is shown once, on a line of
+# its own, both when trace lines are filtered out of it and when they are not.
+check "stderr without a final newline" 0 "report-once:^oops$" \
+  sh -c 'printf oops >&2'
+# shellcheck disable=SC2016 # The expansion is for the traced command to do.
+PS4='+$X '
+export PS4
+check "stderr without a final newline, unfiltered" 0 "report-once:^oops$" \
+  sh -c 'printf oops >&2'
+unset PS4
+
+# Error output that `grep` would treat as binary is still reported.
+check "binary stderr" 0 "report:error$" "$work/binary-stderr"
+
 # Without a command, `cronic` reports a usage error rather than aborting on an
 # unset variable.
 check "no arguments" 64 "message:^Usage: "
@@ -306,42 +369,80 @@ check "caller's exported variables" 0 silent "$work/check-variables" $exported_n
 # shellcheck disable=SC2086  # each name is a separate argument.
 unset $exported_names
 
-# Bash writes its trace lines to a separate file, so they are not error
-# output, whatever PS4 is.
+# A bash that supports BASH_XTRACEFD and the allocation of a file descriptor
+# (bash 4.1 or later) writes its trace lines to a separate file, so they are
+# not error output, whatever PS4 is.  An earlier bash, such as macOS's
+# /bin/bash, writes them to stderr, where `cronic` cannot recognize them when
+# PS4 contains an expansion or an escape, so it reports them as error output.
+# This tests for the features themselves, rather than for the bash version
+# that `cronic` tests for, so that it detects a wrong version test in `cronic`.
+# `cronic` and the bash scripts above all run under the first `bash` in PATH,
+# so that is the bash tested here.  The probe checks whether a trace line
+# reaches the file, rather than whether bash writes to stderr, because bash can
+# write an unrelated warning to stderr, such as for an uninstalled locale.
+if bash -c 'exec {fd}> "$1" && BASH_XTRACEFD=$fd && set -x && :' \
+  bash "$work/xtracefd-probe" 2> /dev/null && [ -s "$work/xtracefd-probe" ]; then
+  bash_has_xtracefd=true
+else
+  bash_has_xtracefd=false
+fi
 export PS4
 # shellcheck disable=SC2016 # The expansion is for the traced command to do.
 PS4='+${LINENO}: '
-check "bash, PS4 with expansions, trace-only stderr" 0 silent \
-  "$work/trace-only" 0
-check "bash, PS4 with expansions, trace in the report" 3 \
-  "report:echo 'the standard output'$" "$work/trace-only" 3
+# The real error is reported, whether or not the trace lines are too.
 check "bash, PS4 with expansions, trace lines and real stderr" 0 \
   "report:^a real error$" "$work/trace-and-stderr" 0
-PS4='+\D{%H}: '
-check "bash, PS4 with an escape, trace-only stderr" 0 silent \
-  "$work/trace-only" 0
+check "bash, PS4 with expansions, trace in the report" 3 \
+  "report:echo 'the standard output'$" "$work/trace-only" 3
+if [ "$bash_has_xtracefd" = true ]; then
+  check "bash, PS4 with expansions, trace-only stderr" 0 silent \
+    "$work/trace-only" 0
+  PS4='+\D{%H}: '
+  check "bash, PS4 with an escape, trace-only stderr" 0 silent \
+    "$work/trace-only" 0
+fi
 PS4='[trace] '
 check "bash, PS4 with metacharacters, trace-only stderr" 0 silent \
   "$work/trace-only" 0
 
-# The tests below concern trace lines that a shell other than bash writes to
-# stderr; they are recognized by PS4.  Bash running as root does not import
-# PS4 from the environment, so `cronic` would see the default PS4, and these
-# tests are skipped.
+# The tests below concern trace lines in stderr, as written by a shell other
+# than bash or by a bash that does not use BASH_XTRACEFD; they are recognized
+# by PS4.  Bash running as root does not import PS4 from the environment, so
+# `cronic` would see the default PS4, and these tests are skipped.
 if [ "$(id -u)" -eq 0 ]; then
   echo "SKIP: tests of PS4 in stderr, because bash ignores PS4 when run as root"
   unset PS4
   exit "$status"
 fi
 
+# An earlier bash writes its trace lines to stderr, and when PS4 contains an
+# expansion or an escape, they are reported.
+if [ "$bash_has_xtracefd" = false ]; then
+  # shellcheck disable=SC2016 # The expansion is for the traced command to do.
+  PS4='+${LINENO}: '
+  check "bash before 4.1, PS4 with expansions, trace-only stderr" 0 \
+    "report:^+3: echo 'the standard output'$" "$work/trace-only" 0
+  PS4='+\D{%H}: '
+  check "bash before 4.1, PS4 with an escape, trace-only stderr" 0 \
+    "report:^+[0-9][0-9]: echo 'the standard output'$" "$work/trace-only" 0
+fi
+
 # Trace lines are recognized when PS4 contains regular-expression
 # metacharacters or quotes, and the metacharacters do not match other lines.
 PS4='[trace] '
 check "PS4 with metacharacters, trace-only stderr" 0 silent \
-  "$work/stderr-lines" "[trace] echo hi" "[[trace] echo nested"
+  "$work/stderr-lines" "[trace] echo hi" "[[trace] echo nested" \
+  "[[[trace] echo nested more"
+# The same, for the trace lines that bash writes to stderr.
+check "bash, PS4 with metacharacters, nested trace in stderr" 0 silent \
+  "$work/nested-trace" 0
+check "bash, PS4 with metacharacters, nested trace in the report" 3 \
+  "report:^\[\[\[trace\] echo deepest$" "$work/nested-trace" 3
 check "PS4 with metacharacters, trace lines and real stderr" 0 \
   "report:^a real error$" \
   "$work/stderr-lines" "[trace] echo hi" "a real error"
+check "PS4 with metacharacters, trace lines and binary stderr" 0 \
+  "report:error$" "$work/binary-stderr"
 PS4='+.* '
 check "PS4 with metacharacters, error line that they would match" 0 \
   "report:^+++ b/file$" "$work/stderr-lines" "+++ b/file"

tests/cronic-test/temp-file-test.sh

@@ -88,11 +88,13 @@ marker="cronic temp-file-test marker $work"
 
 # marker_files: prints the files that contain the marker.  Files directly
 # within /tmp, or within a subdirectory of it, are searched, except for this
-# test's own files.  Errors, such as unreadable files, are ignored.
+# test's own files.  Errors, such as unreadable files, are ignored.  `-H`
+# because on macOS, /tmp is a symbolic link, which `find` otherwise does not
+# follow.
 marker_files() {
   {
     find "$TMPDIR" -type f -exec grep -lF -- "$marker" {} + 2> /dev/null || true
-    find /tmp -maxdepth 2 -path "$work" -prune \
+    find -H /tmp -maxdepth 2 -path "$work" -prune \
       -o -type f -user "$(id -u)" -exec grep -lF -- "$marker" {} + \
       2> /dev/null || true
   } | sort -u