Portability: mostly NetBSD issues - see #553

README.rst

@@ -184,6 +184,7 @@ These perl packages include:
 - Devel::StackTrace
 - Digest::MD5
 - File::Spec
+- GD
 - at least one flavor of JSON module. In order of performance/preference:
 
   - JSON::XS
@@ -213,13 +214,23 @@ To execute ``make install ...`` or ``make doc``, you will need:
 Some of the applications provided with the lcov module are written
 in Python - and may require additional Python packages.
 In particular, 'xlsxwriter' is required in order to generate any
-of the spreadsheet reports, and 'xml' and 'zipfile' are required in
+of the spreadsheet reports, and Python packages 'xml' and 'zipfile' are
+required in
 order to be able to run regression tests.
 
 To measure Python code coverage, users will need Python packages:
 
 - Coverage.py
-
+
+Certain tests use additional tools:
+
+- ``jq`` - the command-line json parser - is required by some tests.
+  ``make check`` fail if it is not found.
+- ``py2lcov`` tests use Coverage.py; these tests will be skipped if
+  the tool is not found in your environment.
+- some LLVM-related tests (*e.g.*, the ``llvm2lcov`` tests) require
+  ``clang`` and/or ``clang++``.  If these are not present in your
+  environment, then the LLVM-related tests are skipped.
 
 In addition, contributors will need:
 

bin/lcov

@@ -136,6 +136,7 @@ sub setup_gkv();
 sub lcov_geninfo(@);
 sub create_package($$$;$);
 sub summary();
+sub count_data_cb($$$);
 
 # Global variables & initialization
 
@@ -1039,6 +1040,22 @@ sub read_file($)
     return $content;
 }
 
+#
+# count_data_cb(dir, relative_name, count_ref)
+#
+# Count the data files found while searching an unpacked package.  Returns
+# undefined:  a defined value would stop 'lcov_find' at the first match.
+#
+
+sub count_data_cb($$$)
+{
+    my ($dir, $rel, $count_ref) = @_;
+
+    $$count_ref++;
+
+    return;
+}
+
 #
 # get_package(package_file)
 #
@@ -1060,16 +1077,16 @@ sub get_package($)
     info("Reading package $file:\n");
     $file = abs_path($file);
     chdir($dir);
-    open(HANDLE, "-|", "tar xvfz " . quotemeta($file) . " 2>$lcovutil::devnull")
+    # Extract quietly:  the format of the verbose listing depends on which
+    #   tar implementation is installed (GNU tar prints just the path, while
+    #   bsdtar appends the file size to each line), so the data files are
+    #   counted in the expanded directory rather than from the tar listing.
+    open(HANDLE, "-|", "tar xfz " . quotemeta($file) . " 2>$lcovutil::devnull")
         or
         die("could not process package $file: $!\n");
-    $count = 0;
-    while (<HANDLE>) {
-        if (/\.da$/ || /\.gcda$/) {
-            $count++;
-        }
-    }
     close(HANDLE);
+    $count = 0;
+    lcov_find($dir, \&count_data_cb, \$count, '\.gcda$', '\.da$');
     if ($count == 0) {
         die("no data file found in package $file\n");
     }
@@ -1126,7 +1143,8 @@ sub count_package_data($)
 
     open(HANDLE, "-|", "tar tfz " . quotemeta($filename)) or return undef;
     while (<HANDLE>) {
-        if (/\.da$/ || /\.gcda$/) {
+        # tolerate 'bsdtar', which appends the file size to each listing line
+        if (/\.(?:gcda|da)(?:\s|\z)/) {
             $count++;
         }
     }

tests/common.tst

@@ -8,6 +8,46 @@ PROFILE="--profile"
 LOCAL_COVERAGE=1
 KEEP_GOING=0
 
+# ---------------------------------------------------------------------------
+# Portable helpers
+# ---------------------------------------------------------------------------
+
+# BSD 'wc -l' left-pads its output, so a literal string comparison against an
+# expected count fails even when the counts are numerically equal.  Always
+# compare '$(count_lines FILE)' - awk prints a clean number everywhere.
+# BSD/macOS has 'md5' (stdout) but not the GNU 'md5sum' that was forked in
+# legacy scripts; compute the identical lowercase hex digest in-process using
+# perl Digest::MD5 so tests agree with the tools on every platform.
+compute_md5_hex()
+{
+    perl -MDigest::MD5 -e '
+        $SIG{PIPE} = q{IGNORE};
+        my $name = @ARGV && $ARGV[0] ne q{-} ? $ARGV[0] : q{-};
+        binmode STDIN;
+        my $self = Digest::MD5->new;
+        $self->add(do { local $/; my $fh = $name eq q{-} ? *STDIN : do { open my $h, q{<}, $name or die qq{md5: $name: $!}; $h }; <$fh> });
+        print $self->hexdigest, chr(10);
+    ' "$@"
+}
+
+count_lines()
+{
+    awk 'END { print NR }' "$@"
+}
+
+# Prints the best way to invoke the python Coverage.py module - empty when
+# the package isn't installed (callers should then SKIP, not fail).
+find_python_coverage()
+{
+    if command -v coverage >/dev/null 2> /dev/null ; then
+        echo "coverage"
+    elif command -v python3-coverage >/dev/null 2> /dev/null ; then
+        echo "python3-coverage"
+    elif python3 -m coverage --version >/dev/null 2> /dev/null ; then
+        echo "python3 -m coverage"
+    fi
+}
+
 #echo "CMD:  $0 $@"
 
 while [ $# -gt 0 ] ; do
@@ -68,14 +108,20 @@ while [ $# -gt 0 ] ; do
                     CMD='python3-coverage' # ubuntu?
                 fi
             fi
-            which $CMD
-            if [ 0 != $? ] ; then
-                echo "cannot find 'coverage' or 'python3-coverage'"
-                echo "unable to run py2lcov - please install python Coverage.py package"
-                exit 1
+            if ! command -v "$CMD" > /dev/null 2> /dev/null ; then
+                if [ "python3 -m coverage" != "$CMD" ] &&
+                   ! python3 -m coverage --version >/dev/null 2> /dev/null ; then
+                    # Not a bug in the tools - run without coverage
+                    #   instrumentation (see runtests.py: a missing Coverage.py
+                    #   also counts as a platform skip there).
+                    echo "python Coverage.py package not installed - running without runtime instrumentation"
+                    CMD=''
+                fi
             fi
 
-            PYCOVER="COVERAGE_FILE=$PYCOV_DB $CMD run --branch --append"
+            if [ '' != "$CMD" ] ; then
+                PYCOVER="COVERAGE_FILE=$PYCOV_DB $CMD run --branch --append"
+            fi
             ;;
 
         --home | -home )
@@ -266,7 +312,7 @@ function check_tla_css()
     local rc=0
     local class
     for class in `find $dir -name '*.html' -print0 |
-                  xargs -0 --no-run-if-empty grep -h -o -E 'class="[^"]+"' |
+                  xargs -0 -r grep -h -o -E 'class="[^"]+"' |
                   sed -e 's/class="//' -e 's/"$//' | tr ' ' '\n' |
                   grep -E '^tla(Bg)?[A-Z]' | sort -u` ; do
         if ! grep -E -q "(^|[[:space:]])(td|span|a)\.$class([^A-Za-z0-9]|\$)" \

tests/genhtml/htmlgen/htmlgen.sh

@@ -430,8 +430,8 @@ for mode in plain nav ; do
     done
     # the source text has no brackets in it, so every '[' in the page opened a
     #   coverpoint group and every ']' closed one
-    OPEN=`grep -o '\[' $PAGE | wc -l`
-    CLOSE=`grep -o '\]' $PAGE | wc -l`
+    OPEN=`grep -o '\[' $PAGE | grep -c ''`
+    CLOSE=`grep -o '\]' $PAGE | grep -c ''`
     if [ "$OPEN" != "$CLOSE" ] ; then
         fail unreach_$mode "$OPEN '[' but $CLOSE ']' in the source page"
     fi
@@ -448,7 +448,7 @@ for mode in plain nav ; do
     #   matches and not lines - and the page is found first, because backticks
     #   inside backticks do not nest
     PAGE=`find html_$mode -name 't.c.gcov.html'`
-    COUNT=`grep -o -F -e 'tlaEUC' $PAGE | wc -l`
+    COUNT=`grep -o -F -e 'tlaEUC' $PAGE | grep -c ''`
     eval "COUNT_$mode=$COUNT"
 done
 if [ "$COUNT_plain" != "$COUNT_nav" ] ; then
@@ -817,7 +817,7 @@ for mode in nfb nfb_off nfb_noann nfb_olderbase ; do
     # the MC/DC coverpoints are on the page at all:  four markers on line 5,
     #   which are dropped entirely if the categorizer walks only the baseline.
     #   All four are on one line, so this counts matches and not lines
-    COUNT=`grep -o -F -e 'class="mcdcTla"' $PAGE | wc -l`
+    COUNT=`grep -o -F -e 'class="mcdcTla"' $PAGE | grep -c ''`
     if [ 4 != "$COUNT" ] ; then
         fail $mode "expected 4 MC/DC markers in the $mode page, found $COUNT"
     fi

tests/genpng/genpng.sh

@@ -183,7 +183,7 @@ elif [ -s genpng2.log ] ; then
 else
     dump hand.png > hand.dump
     # nine source lines, and no row for either of the two metadata lines
-    ROWS=`wc -l < hand.dump`
+    ROWS=`count_lines hand.dump`
     if [ "$ROWS" != 9 ] ; then
         fail "Test 2 layout: expected 9 rows, got $ROWS: `cat hand.dump`"
     else
@@ -230,7 +230,7 @@ if [ 0 != ${PIPESTATUS[0]} ] ; then
     fail "Test 3 plain text: genpng failed"
 else
     dump plain.png > plain.dump
-    if [ "`wc -l < plain.dump`" != 3 ] ; then
+    if [ "`count_lines plain.dump`" != 3 ] ; then
         fail "Test 3 plain text: expected 3 rows: `cat plain.dump`"
     elif [ "`grep -c \" $PLAIN \" plain.dump`" != 3 ] ; then
         fail "Test 3 plain text: not every row is plain: `cat plain.dump`"

tests/lcov/extract/extract1.sh

@@ -181,7 +181,13 @@ if [ 0 != $? ] ; then
 fi
 
 # exclude some un-covered standard includes
-EXCLUDE='--exclude */include/c++/* --exclude */include/g++-v* --ignore unused'
+# Depending on platform, the compiler's private headers live under
+# '/usr/include/c++/N' (Linux), '/usr/include/g++/bits' or
+# '/usr/pkg/include/gcc/N' - exclude every spelling, tolerating the unused
+# ones.
+EXCLUDE='--exclude */include/c++/* --exclude */include/g++-v*
+        --exclude */include/g++/* --exclude */include/gcc/*
+        --exclude */include/gcc[0-9]*/* --ignore unused'
 
 # callback tests
 echo $COVER $CAPTURE . $LCOV_OPTS -o callback.info $FILTER $IGNORE --criteria $SCRIPT_DIR/threshold.pm,--line,90,--branch,65,--function,100 $EXCLUDE

tests/lcov/extract/extract2.sh

@@ -761,7 +761,7 @@ fi
 
 # test filename containing spaces
 rm -rf ./mytest
-mkdir -pv ./mytest
+mkdir -p ./mytest
 echo "int main(){}" > './mytest/main space.cpp'
 ( cd ./mytest ; ${CXX} -c  'main space.cpp' --coverage )
 

tests/lcov/fork_throttle/fork_throttle.pl

@@ -28,6 +28,22 @@
 
 lcovutil::parseOptions({}, {});
 
+# The throttle's whole subject is a memory governor:  it learns a byte-per-
+#   unit rate from a child's peak RSS, which is only measurable where the OS
+#   reports one - the same fact which makes lcovutil emit no per-job
+#   'memory' profile data at all on platforms which don't expose it (their
+#   read_proc_peak_memory returns zeroes).  Without that signal every check
+#   below is moot - and worse, the 'learnedBytes / learnedWeight' rate
+#   divides by a never-learned weight, which is the crash this guard exists
+#   to prevent - so skip rather than fail.
+{
+    my ($rss, $vsize) = lcovutil::read_proc_peak_memory();
+    if (!$rss && !$vsize) {
+        print("skipped: platform does not report peak memory\n");
+        exit(0);
+    }
+}
+
 my $tempdir = 'fork_throttle.tmp';
 rmtree($tempdir) if -d $tempdir;
 mkdir($tempdir) or die("cannot create $tempdir: $!");

tests/lcov/parallel_parse/parallel_parse.sh

@@ -372,7 +372,7 @@ fi
 # 9. truncated final section:  the same format error as a serial read, with
 #    the same line numbers
 #-----------------------------------------------------------------------
-head -n -1 multi.info > trunc.info
+sed '$ d' multi.info > trunc.info
 run_pair trunc trunc.info --ignore-errors corrupt,format
 for f in trunc_serial.log trunc_split.log ; do
     if ! grep -E "unexpected end of file: missing 'end_of_record'" $f \

tests/lcov/scheduling/scheduling.sh

@@ -323,16 +323,22 @@ check_memory agg_prof.json filter filt_child .filt_child
 check_memory agg_prof.json aggregate 'group total' \
     '(with_entries(select(.key | test("^[0-9]+$"))))'
 
-# Both phases must be present simultaneously in that one profile:  under the
-# earlier pid keying they were indistinguishable, and under a flat numeric
-# keying they would have overwritten each other.
-for phase in filter aggregate ; do
-    n=`jq -r --arg p "$phase" '[.memory | keys[]
-           | select(startswith($p + "_"))] | length' agg_prof.json`
-    if [ "${n:-0}" -lt 1 ] ; then
-        fail_memory $phase "no $phase worker in agg_prof.json"
-    fi
-done
+if [ "`jq -r 'has("memory")' agg_prof.json`" != "true" ] ; then
+    echo "memory data absent in agg_prof.json (platform does not expose peak memory)"
+else
+    {
+    # Both phases must be present simultaneously in that one profile:  under the
+    # earlier pid keying they were indistinguishable, and under a flat numeric
+    # keying they would have overwritten each other.
+    for phase in filter aggregate ; do
+        n=`jq -r --arg p "$phase" '[.memory | keys[]
+               | select(startswith($p + "_"))] | length' agg_prof.json`
+        if [ "${n:-0}" -lt 1 ] ; then
+            fail_memory $phase "no $phase worker in agg_prof.json"
+        fi
+    done
+    }
+fi
 # ...and a collision would have been reported by merge_child_profile
 if grep -i 'unexpected duplicate key' agg.log geninfo_hist.log genhtml_hist.log ; then
     fail_memory collision \
@@ -364,13 +370,19 @@ if grep -i 'unexpected duplicate key' nested.log ; then
 fi
 # the qualified filter workers, and the check that memory{filter_<id>} still
 # lines up with filt_child{<id>} for them
-nested=`jq -r '[.memory | keys[] | select(test("^filter_aggregate_"))]
-               | sort | join(" ")' nested_prof.json`
-if [ "`echo $nested | wc -w`" -lt 2 ] ; then
-    jq -r '.memory | keys' nested_prof.json
-    fail_memory nested "no nested filter worker in nested_prof.json"
+if [ "`jq -r 'has("memory")' nested_prof.json`" != "true" ] ; then
+    echo "memory data absent in nested_prof.json (platform does not expose peak memory)"
 else
-    echo "OK (memory nested): [$nested]"
+    {
+    nested=`jq -r '[.memory | keys[] | select(test("^filter_aggregate_"))]
+                   | sort | join(" ")' nested_prof.json`
+    if [ "`echo $nested | wc -w`" -lt 2 ] ; then
+        jq -r '.memory | keys' nested_prof.json
+        fail_memory nested "no nested filter worker in nested_prof.json"
+    else
+        echo "OK (memory nested): [$nested]"
+    fi
+    }
 fi
 check_memory nested_prof.json filter filt_child .filt_child
 

tests/py2lcov/py2lcov.sh

@@ -51,17 +51,14 @@ LCOV_OPTS="--branch-coverage $PARALLEL $PROFILE"
 if [ '' != "${COVERAGE_COMMAND}" ] ; then
     CMD=${COVERAGE_COMMAND}
 else
-    CMD='coverage'
-    which $CMD
-    if [ 0 != $? ] ; then
-        CMD='python3-coverage' # ubuntu?
-    fi
+    CMD=$(find_python_coverage)
 fi
-which $CMD
-if [ 0 != $? ] ; then
-    echo "cannot find 'coverage' or 'python3-coverage'"
-    echo "unable to run py2lcov - please install python Coverage.py package"
-    exit 1
+if [ '' == "$CMD" ] ; then
+    # The python Coverage.py package is not installed on this platform:
+    #   py2lcov itself is not implicated - skip instead of failing
+    #   (same treatment as other platform-constraint skips in this suite).
+    echo "python Coverage.py package not installed - skipping py2lcov test"
+    exit 0
 fi
 
 # some corner cases: