WIP Unknown progress, saving for reference

Inspired by Linux's Bloat-O-Meter, code_size.py wraps nm to provide
function-level code size, and supports detailed comparison between
different builds.

One difference is that code_size.py invokes littlefs's build system
similarly to test.py, creating a duplicate build in the "sizes"
directory. This makes it easy to monitor a cross-compiled build size
while simultaneously testing on the host machine.

Makefile

@@ -6,6 +6,7 @@ endif
 CC ?= gcc
 AR ?= ar
 SIZE ?= size
+NM ?= nm
 
 SRC += $(wildcard *.c bd/*.c)
 OBJ := $(SRC:.c=.o)
@@ -29,6 +30,7 @@ override CFLAGS += -Wextra -Wshadow -Wjump-misses-init -Wundef
 
 ifdef VERBOSE
 override TFLAGS += -v
+override SFLAGS += -v
 endif
 
 
@@ -39,6 +41,9 @@ asm: $(ASM)
 size: $(OBJ)
 	$(SIZE) -t $^
 
+code_size:
+	./scripts/code_size.py $(SFLAGS)
+
 test:
 	./scripts/test.py $(TFLAGS)
 .SECONDEXPANSION:
@@ -65,3 +70,4 @@ clean:
 	rm -f $(DEP)
 	rm -f $(ASM)
 	rm -f tests/*.toml.*
+	rm -f sizes/*

bd/lfs_filebd.c

@@ -80,11 +80,6 @@ int lfs_filebd_read(const struct lfs_config *cfg, lfs_block_t block,
     LFS_ASSERT(size % cfg->read_size == 0);
     LFS_ASSERT(block < cfg->block_count);
 
-    // zero for reproducability (in case file is truncated)
-    if (bd->cfg->erase_value != -1) {
-        memset(buffer, bd->cfg->erase_value, size);
-    }
-
     // read
     off_t res1 = lseek(bd->fd,
             (off_t)block*cfg->block_size + (off_t)off, SEEK_SET);
@@ -101,6 +96,11 @@ int lfs_filebd_read(const struct lfs_config *cfg, lfs_block_t block,
         return err;
     }
 
+    // file truncated? zero for reproducability
+    if ((lfs_size_t)res2 < size) {
+        memset((uint8_t*)buffer + res2, 0, size-res2);
+    }
+
     LFS_FILEBD_TRACE("lfs_filebd_read -> %d", 0);
     return 0;
 }

bd/lfs_rambd.c

@@ -32,11 +32,8 @@ int lfs_rambd_createcfg(const struct lfs_config *cfg,
         }
     }
 
-    // zero for reproducability?
+    // zero for reproducability (this matches filebd)
-    if (bd->cfg->erase_value != -1) {
+    memset(bd->buffer, 0, cfg->block_size * cfg->block_count);
-        memset(bd->buffer, bd->cfg->erase_value,
-                cfg->block_size * cfg->block_count);
-    }
 
     LFS_RAMBD_TRACE("lfs_rambd_createcfg -> %d", 0);
     return 0;

lfs.h

@@ -85,6 +85,7 @@ enum lfs_error {
     LFS_ERR_NOMEM       = -12,  // No more memory available
     LFS_ERR_NOATTR      = -61,  // No data/attr available
     LFS_ERR_NAMETOOLONG = -36,  // File name too long
+    LFS_ERR_OVERFLOW    = -75,  // Value too large for defined data type
 };
 
 // File types
@@ -113,11 +114,20 @@ enum lfs_type {
     LFS_TYPE_SOFTTAIL       = 0x600,
     LFS_TYPE_HARDTAIL       = 0x601,
     LFS_TYPE_MOVESTATE      = 0x7ff,
+    LFS_TYPE_COMMITCRC      = 0x502,
+    LFS_TYPE_NPROGCRC       = 0x5ff,
 
     // internal chip sources
     LFS_FROM_NOOP           = 0x000,
     LFS_FROM_MOVE           = 0x101,
     LFS_FROM_USERATTRS      = 0x102,
+
+    // internally used tag-types
+    LFS_T_CRC               = 0xc,
+    LFS_T_USERATTR          = 0x2,
+
+    // internally used tag-subtypes
+    LFS_S_NPROG             = 0x01,
 };
 
 // File open flags
@@ -319,6 +329,20 @@ typedef struct lfs_cache {
     uint8_t *buffer;
 } lfs_cache_t;
 
+typedef struct lfs_mdir_ {
+    lfs_block_t pair[2];
+    uint32_t rev;
+    lfs_off_t roff;
+    lfs_off_t eoff;
+    uint16_t count;
+
+    // TODO need both erased and eperturb?
+    bool erased;
+    bool split;
+    bool eperturb;
+    lfs_block_t tail[2];
+} lfs_mdir__t;
+
 typedef struct lfs_mdir {
     lfs_block_t pair[2];
     uint32_t rev;

scripts/code_size.py

@@ -0,0 +1,328 @@
+#!/usr/bin/env python3
+#
+# This script finds the code size at the function level, with/without
+# static functions, and has some conveniences for comparing different
+# versions. It's basically one big wrapper around nm, and may or may
+# not have been written out of jealousy of Linux's Bloat-O-Meter.
+#
+# Here's a useful bash script to use while developing:
+# ./scripts/code_size.py -qo old.csv
+# while true ; do ./code_scripts/size.py -d old.csv ; inotifywait -rqe modify * ; done
+#
+# Or even better, to automatically update results on commit:
+# ./scripts/code_size.py -qo commit.csv
+# while true ; do ./scripts/code_size.py -d commit.csv -o current.csv ; git diff --exit-code --quiet && cp current.csv commit.csv ; inotifywait -rqe modify * ; done
+#
+# Or my personal favorite:
+# ./scripts/code_size.py -qo master.csv && cp master.csv commit.csv
+# while true ; do ( ./scripts/code_size.py -i commit.csv -d master.csv -s ; ./scripts/code_size.py -i current.csv -d master.csv -s ; ./scripts/code_size.py -d master.csv -o current.csv -s ) | awk 'BEGIN {printf "%-16s %7s %7s %7s\n","","old","new","diff"} (NR==2 && $1="commit") || (NR==4 && $1="prev") || (NR==6 && $1="current") {printf "%-16s %7s %7s %7s %s\n",$1,$2,$3,$5,$6}' ; git diff --exit-code --quiet && cp current.csv commit.csv ; inotifywait -rqe modify * ; done
+#
+
+import os
+import itertools as it
+import subprocess as sp
+import shlex
+import re
+import csv
+import collections as co
+
+SIZEDIR = 'sizes'
+RULES = """
+define FLATTEN
+%(sizedir)s/%(build)s.$(subst /,.,$(target)): $(target)
+    ( echo "#line 1 \\"$$<\\"" ; %(cat)s $$< ) > $$@
+%(sizedir)s/%(build)s.$(subst /,.,$(target:.c=.size)): \\
+        %(sizedir)s/%(build)s.$(subst /,.,$(target:.c=.o))
+    $(NM) --size-sort $$^ | sed 's/^/$(subst /,\\/,$(target:.c=.o)):/' > $$@
+endef
+$(foreach target,$(SRC),$(eval $(FLATTEN)))
+
+-include %(sizedir)s/*.d
+.SECONDARY:
+
+%%.size: $(foreach t,$(subst /,.,$(SRC:.c=.size)),%%.$t)
+    cat $^ > $@
+"""
+CATS = {
+    'code': 'cat',
+    'code_inlined': 'sed \'s/^static\( inline\)\?//\'',
+}
+
+def build(**args):
+    # mkdir -p sizedir
+    os.makedirs(args['sizedir'], exist_ok=True)
+
+    if args.get('inlined', False):
+        builds = ['code', 'code_inlined']
+    else:
+        builds = ['code']
+
+    # write makefiles for the different types of builds
+    makefiles = []
+    targets = []
+    for build in builds:
+        path = args['sizedir'] + '/' + build
+        with open(path + '.mk', 'w') as mk:
+            mk.write(RULES.replace(4*' ', '\t') % dict(
+                sizedir=args['sizedir'],
+                build=build,
+                cat=CATS[build]))
+            mk.write('\n')
+
+            # pass on defines
+            for d in args['D']:
+                mk.write('%s: override CFLAGS += -D%s\n' % (
+                    path+'.size', d))
+
+        makefiles.append(path + '.mk')
+        targets.append(path + '.size')
+
+    # build in parallel
+    cmd = (['make', '-f', 'Makefile'] +
+        list(it.chain.from_iterable(['-f', m] for m in makefiles)) +
+        [target for target in targets])
+    if args.get('verbose', False):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.DEVNULL if not args.get('verbose', False) else None)
+    proc.wait()
+    if proc.returncode != 0:
+        sys.exit(-1)
+
+    # find results
+    build_results = co.defaultdict(lambda: 0)
+    # notes
+    # - filters type
+    # - discards internal/debug functions (leading __)
+    pattern = re.compile(
+        '^(?P<file>[^:]+)' +
+        ':(?P<size>[0-9a-fA-F]+)' +
+        ' (?P<type>[%s])' % re.escape(args['type']) +
+        ' (?!__)(?P<name>.+?)$')
+    for build in builds:
+        path = args['sizedir'] + '/' + build
+        with open(path + '.size') as size:
+            for line in size:
+                match = pattern.match(line)
+                if match:
+                    file = match.group('file')
+                    # discard .8449 suffixes created by optimizer
+                    name = re.sub('\.[0-9]+', '', match.group('name'))
+                    size = int(match.group('size'), 16)
+                    build_results[(build, file, name)] += size
+
+    results = []
+    for (build, file, name), size in build_results.items():
+        if build == 'code':
+            results.append((file, name, size, False))
+        elif (build == 'code_inlined' and
+                ('inlined', file, name) not in results):
+            results.append((file, name, size, True))
+
+    return results
+
+def main(**args):
+    # find results
+    if not args.get('input', None):
+        results = build(**args)
+    else:
+        with open(args['input']) as f:
+            r = csv.DictReader(f)
+            results = [
+                (   result['file'],
+                    result['name'],
+                    int(result['size']),
+                    bool(int(result.get('inlined', 0))))
+                for result in r
+                if (not bool(int(result.get('inlined', 0))) or
+                    args.get('inlined', False))]
+
+    total = 0
+    for _, _, size, inlined in results:
+        if not inlined:
+            total += size
+
+    # find previous results?
+    if args.get('diff', None):
+        with open(args['diff']) as f:
+            r = csv.DictReader(f)
+            prev_results = [
+                (   result['file'],
+                    result['name'],
+                    int(result['size']),
+                    bool(int(result.get('inlined', 0))))
+                for result in r
+                if (not bool(int(result.get('inlined', 0))) or
+                    args.get('inlined', False))]
+
+        prev_total = 0
+        for _, _, size, inlined in prev_results:
+            if not inlined:
+                prev_total += size
+
+    # write results to CSV
+    if args.get('output', None):
+        results.sort(key=lambda x: (-x[2], x))
+        with open(args['output'], 'w') as f:
+            w = csv.writer(f)
+            if args.get('inlined', False):
+                w.writerow(['file', 'name', 'size', 'inlined'])
+                for file, name, size, inlined in results:
+                    w.writerow((file, name, size, int(inlined)))
+            else:
+                w.writerow(['file', 'name', 'size'])
+                for file, name, size, inlined in results:
+                    w.writerow((file, name, size))
+
+    # print results
+    def dedup_functions(results):
+        functions = co.defaultdict(lambda: (0, True))
+        for _, name, size, inlined in results:
+            if not inlined:
+                functions[name] = (functions[name][0] + size, False)
+        for _, name, size, inlined in results:
+            if inlined and functions[name][1]:
+                functions[name] = (functions[name][0] + size, True)
+        return functions
+
+    def dedup_files(results):
+        files = co.defaultdict(lambda: 0)
+        for file, _, size, inlined in results:
+            if not inlined:
+                files[file] += size
+        return files
+
+    def diff_sizes(olds, news):
+        diff = co.defaultdict(lambda: (None, None, None))
+        for name, new in news.items():
+            diff[name] = (None, new, new)
+        for name, old in olds.items():
+            new = diff[name][1] or 0
+            diff[name] = (old, new, new-old)
+        return diff
+
+    def print_header(name=''):
+        if not args.get('diff', False):
+            print('%-40s %7s' % (name, 'size'))
+        else:
+            print('%-40s %7s %7s %7s' % (name, 'old', 'new', 'diff'))
+
+    def print_functions():
+        functions = dedup_functions(results)
+        functions = {
+            name+' (inlined)' if inlined else name: size
+            for name, (size, inlined) in functions.items()}
+
+        if not args.get('diff', None):
+            print_header('function')
+            for name, size in sorted(functions.items(),
+                    key=lambda x: (-x[1], x)):
+                print("%-40s %7d" % (name, size))
+        else:
+            prev_functions = dedup_functions(prev_results)
+            prev_functions = {
+                name+' (inlined)' if inlined else name: size
+                for name, (size, inlined) in prev_functions.items()}
+            diff = diff_sizes(functions, prev_functions)
+            print_header('function (%d added, %d removed)' % (
+                sum(1 for old, _, _ in diff.values() if not old),
+                sum(1 for _, new, _ in diff.values() if not new)))
+            for name, (old, new, diff) in sorted(diff.items(),
+                    key=lambda x: (-(x[1][2] or 0), x)):
+                if diff or args.get('all', False):
+                    print("%-40s %7s %7s %+7d%s" % (
+                        name, old or "-", new or "-", diff,
+                        ' (%+.2f%%)' % (100*((new-old)/old))
+                        if old and new else
+                        ''))
+
+    def print_files():
+        files = dedup_files(results)
+
+        if not args.get('diff', None):
+            print_header('file')
+            for file, size in sorted(files.items(),
+                    key=lambda x: (-x[1], x)):
+                print("%-40s %7d" % (file, size))
+        else:
+            prev_files = dedup_files(prev_results)
+            diff = diff_sizes(files, prev_files)
+            print_header('file (%d added, %d removed)' % (
+                sum(1 for old, _, _ in diff.values() if not old),
+                sum(1 for _, new, _ in diff.values() if not new)))
+            for name, (old, new, diff) in sorted(diff.items(),
+                    key=lambda x: (-(x[1][2] or 0), x)):
+                if diff or args.get('all', False):
+                    print("%-40s %7s %7s %+7d%s" % (
+                        name, old or "-", new or "-", diff,
+                        ' (%+.2f%%)' % (100*((new-old)/old))
+                        if old and new else
+                        ''))
+
+    def print_totals():
+        if not args.get('diff', None):
+            print("%-40s %7d" % ('TOTALS', total))
+        else:
+            print("%-40s %7s %7s %+7d%s" % (
+                'TOTALS', prev_total, total, total-prev_total,
+                ' (%+.2f%%)' % (100*((total-prev_total)/total))
+                if prev_total and total else
+                ''))
+
+    def print_status():
+        if not args.get('diff', None):
+            print(total)
+        else:
+            print("%d (%+.2f%%)" % (total, 100*((total-prev_total)/total)))
+
+    if args.get('quiet', False):
+        pass
+    elif args.get('status', False):
+        print_status()
+    elif args.get('summary', False):
+        print_header()
+        print_totals()
+    elif args.get('files', False):
+        print_files()
+        print_totals()
+    else:
+        print_functions()
+        print_totals()
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+    parser = argparse.ArgumentParser(
+        description="Find code size at the function level.")
+    parser.add_argument('sizedir', nargs='?', default=SIZEDIR,
+        help="Directory to store intermediary results. Defaults "
+            "to \"%s\"." % SIZEDIR)
+    parser.add_argument('-D', action='append', default=[],
+        help="Specify compile-time define.")
+    parser.add_argument('-v', '--verbose', action='store_true',
+        help="Output commands that run behind the scenes.")
+    parser.add_argument('-i', '--input',
+        help="Don't compile and find code sizes, instead use this CSV file.")
+    parser.add_argument('-o', '--output',
+        help="Specify CSV file to store results.")
+    parser.add_argument('-d', '--diff',
+        help="Specify CSV file to diff code size against.")
+    parser.add_argument('-a', '--all', action='store_true',
+        help="Show all functions, not just the ones that changed.")
+    parser.add_argument('--inlined', action='store_true',
+        help="Run a second compilation to find the sizes of functions normally "
+            "removed by optimizations. These will be shown as \"*.inlined\" "
+            "functions, and will not be included in the total.")
+    parser.add_argument('--files', action='store_true',
+        help="Show file-level code sizes. Note this does not include padding! "
+            "So sizes may differ from other tools.")
+    parser.add_argument('-s', '--summary', action='store_true',
+        help="Only show the total code size.")
+    parser.add_argument('-S', '--status', action='store_true',
+        help="Show minimum info useful for a single-line status.")
+    parser.add_argument('-q', '--quiet', action='store_true',
+        help="Don't show anything, useful with -o.")
+    parser.add_argument('--type', default='tTrRdDbB',
+        help="Type of symbols to report, this uses the same single-character "
+            "type-names emitted by nm. Defaults to %(default)r.")
+    sys.exit(main(**vars(parser.parse_args())))

scripts/readmdir.py

@@ -24,6 +24,7 @@ TAG_TYPES = {
     'gstate':       (0x700, 0x700),
     'movestate':    (0x7ff, 0x7ff),
     'crc':          (0x700, 0x500),
+    'nprogcrc':     (0x7ff, 0x5ff),
 }
 
 class Tag:
@@ -99,7 +100,16 @@ class Tag:
         return struct.unpack('b', struct.pack('B', self.chunk))[0]
 
     def is_(self, type):
-        return (self.type & TAG_TYPES[type][0]) == TAG_TYPES[type][1]
+        try:
+            if ' ' in type:
+                type1, type3 = type.split()
+                return (self.is_(type1) and
+                    (self.type & ~TAG_TYPES[type1][0]) == int(type3, 0))
+
+            return self.type == int(type, 0)
+
+        except (ValueError, KeyError):
+            return (self.type & TAG_TYPES[type][0]) == TAG_TYPES[type][1]
 
     def mkmask(self):
         return Tag(
@@ -109,14 +119,20 @@ class Tag:
 
     def chid(self, nid):
         ntag = Tag(self.type, nid, self.size)
-        if hasattr(self, 'off'):  ntag.off  = self.off
+        if hasattr(self, 'off'):    ntag.off    = self.off
-        if hasattr(self, 'data'): ntag.data = self.data
+        if hasattr(self, 'data'):   ntag.data   = self.data
-        if hasattr(self, 'crc'):  ntag.crc  = self.crc
+        if hasattr(self, 'crc'):    ntag.crc    = self.crc
+        if hasattr(self, 'erased'): ntag.erased = self.erased
         return ntag
 
     def typerepr(self):
-        if self.is_('crc') and getattr(self, 'crc', 0xffffffff) != 0xffffffff:
+        if (self.is_('crc') and not self.is_('nprogcrc') and
-            return 'crc (bad)'
+                getattr(self, 'crc', 0xffffffff) != 0xffffffff):
+            crc_status = ' (bad)'
+        elif self.is_('nprogcrc') and getattr(self, 'erased', False):
+            crc_status = ' (era)'
+        else:
+            crc_status = ''
 
         reverse_types = {v: k for k, v in TAG_TYPES.items()}
         for prefix in range(12):
@@ -124,12 +140,12 @@ class Tag:
             if (mask, self.type & mask) in reverse_types:
                 type = reverse_types[mask, self.type & mask]
                 if prefix > 0:
-                    return '%s %#0*x' % (
+                    return '%s %#x%s' % (
-                        type, prefix//4, self.type & ((1 << prefix)-1))
+                        type, self.type & ((1 << prefix)-1), crc_status)
                 else:
-                    return type
+                    return '%s%s' % (type, crc_status)
         else:
-            return '%02x' % self.type
+            return '%02x%s' % (self.type, crc_status)
 
     def idrepr(self):
         return repr(self.id) if self.id != 0x3ff else '.'
@@ -172,6 +188,8 @@ class MetadataPair:
 
         self.rev, = struct.unpack('<I', block[0:4])
         crc = binascii.crc32(block[0:4])
+        etag = None
+        estate = None
 
         # parse tags
         corrupt = False
@@ -182,11 +200,11 @@ class MetadataPair:
         while len(block) - off >= 4:
             ntag, = struct.unpack('>I', block[off:off+4])
 
-            tag = Tag(int(tag) ^ ntag)
+            tag = Tag((int(tag) ^ ntag) & 0x7fffffff)
             tag.off = off + 4
             tag.data = block[off+4:off+tag.dsize]
-            if tag.is_('crc'):
+            if tag.is_('crc') and not tag.is_('nprogcrc'):
-                crc = binascii.crc32(block[off:off+4+4], crc)
+                crc = binascii.crc32(block[off:off+2*4], crc)
             else:
                 crc = binascii.crc32(block[off:off+tag.dsize], crc)
             tag.crc = crc
@@ -194,16 +212,29 @@ class MetadataPair:
 
             self.all_.append(tag)
 
-            if tag.is_('crc'):
+            if tag.is_('nprogcrc') and len(tag.data) == 8:
+                etag = tag
+                estate = struct.unpack('<II', tag.data)
+            elif tag.is_('crc'):
                 # is valid commit?
                 if crc != 0xffffffff:
                     corrupt = True
                 if not corrupt:
                     self.log = self.all_.copy()
+                    # end of commit?
+                    if estate:
+                        esize, ecrc = estate
+                        dcrc = 0xffffffff ^ binascii.crc32(block[off:off+esize])
+                        if ecrc == dcrc:
+                            etag.erased = True
+                            corrupt = True
+                    elif not (tag.is_('crc 0x0') or tag.is_('crc 0x1')):
+                        corrupt = True
 
                 # reset tag parsing
                 crc = 0
-                tag = Tag(int(tag) ^ ((tag.type & 1) << 31))
+                etag = None
+                estate = None
 
         # find active ids
         self.ids = list(it.takewhile(
@@ -280,7 +311,7 @@ class MetadataPair:
         f.write('\n')
 
         for tag in tags:
-            f.write("%08x: %08x  %-13s %4s %4s" % (
+            f.write("%08x: %08x  %-14s %3s %4s" % (
                 tag.off, tag,
                 tag.typerepr(), tag.idrepr(), tag.sizerepr()))
             if truncate:

tests/test_mdirs.toml

@@ -0,0 +1,51 @@
+# Test internal operations of metadata-pairs
+
+[[case]] 
+in = "lfs.c"
+code = '''
+    lfs_init(&lfs, &cfg) => 0;
+    lfs_mdir__t mdir = {
+        .pair = {0, 1},
+        .rev = 0,
+        .roff = 0,
+        .eoff = 0,
+        .count = 0,
+        .split = false,
+        .eperturb = false,
+        .tail = {0xffffffff, 0xffffffff},
+    };
+
+
+    struct lfs_mattr_ attrs[] = {
+        {LFS_MKTAG_(LFS_T_USERATTR,  0, 0), "0000", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  1, 0), "0001", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  2, 0), "0002", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  3, 0), "0003", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  4, 0), "0004", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  5, 0), "0005", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  6, 0), "0006", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  7, 0), "0007", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  8, 0), "0008", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR,  9, 0), "0009", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR, 10, 0), "0010", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR, 11, 0), "0011", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR, 12, 0), "0012", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR, 13, 0), "0013", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR, 14, 0), "0014", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR, 15, 0), "0015", 4},
+        {LFS_MKTAG_(LFS_T_USERATTR, 16, 0), "0016", 4}
+    };
+    lfs_size_t attr_count = sizeof(attrs)/sizeof(attrs[0]);
+
+    lfs_dir_compact_(&lfs, &mdir, attrs, attr_count) => 0;
+
+    for (lfs_size_t i = 0; i < attr_count; i++) {
+        printf("looking up %08x\n", attrs[i].tag);
+        lfs_size_t nsize;
+        lfs_dir_getattr_(&lfs, &mdir,
+                attrs[i].tag,
+                buffer, sizeof(buffer), &nsize) => attrs[i].tag;
+        assert(nsize == attrs[i].size);
+        assert(memcmp(buffer, attrs[i].buffer, attrs[i].size) == 0);
+    }
+'''

tests/test_powerloss.toml

@@ -0,0 +1,172 @@
+# There are already a number of tests that test general operations under
+# power-loss (see the reentrant attribute). These tests are for explicitly
+# testing specific corner cases.
+
+[[case]] # only a revision count
+code = '''
+    lfs_format(&lfs, &cfg) => 0;
+
+    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mkdir(&lfs, "notebook") => 0;
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
+    strcpy((char*)buffer, "hello");
+    size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    char rbuffer[256];
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // get pair/rev count
+    lfs_mount(&lfs, &cfg) => 0;
+    lfs_dir_open(&lfs, &dir, "notebook") => 0;
+    lfs_block_t pair[2] = {dir.m.pair[0], dir.m.pair[1]};
+    uint32_t rev = dir.m.rev;
+    lfs_dir_close(&lfs, &dir) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // write just the revision count
+    uint8_t bbuffer[LFS_BLOCK_SIZE];
+    cfg.read(&cfg, pair[1], 0, bbuffer, LFS_BLOCK_SIZE) => 0;
+
+    memcpy(bbuffer, &(uint32_t){lfs_tole32(rev+1)}, sizeof(uint32_t));
+
+    cfg.erase(&cfg, pair[1]) => 0;
+    cfg.prog(&cfg, pair[1], 0, bbuffer, LFS_BLOCK_SIZE) => 0;
+
+    lfs_mount(&lfs, &cfg) => 0;
+
+    // can read?
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    // can write?
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_APPEND) => 0;
+    strcpy((char*)buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    strcpy((char*)buffer, "hello");
+    size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    strcpy((char*)buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[[case]] # partial prog, may not be byte in order!
+if = "LFS_PROG_SIZE < LFS_BLOCK_SIZE"
+define.BYTE_OFF = ["0", "LFS_PROG_SIZE-1", "LFS_PROG_SIZE/2"]
+define.BYTE_VALUE = [0x33, 0xcc]
+in = "lfs.c"
+code = '''
+    lfs_format(&lfs, &cfg) => 0;
+
+    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mkdir(&lfs, "notebook") => 0;
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
+    strcpy((char*)buffer, "hello");
+    size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    char rbuffer[256];
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // imitate a partial prog, value should not matter, if littlefs
+    // doesn't notice the partial prog testbd will assert
+
+    // get offset to next prog
+    lfs_mount(&lfs, &cfg) => 0;
+    lfs_dir_open(&lfs, &dir, "notebook") => 0;
+    lfs_block_t block = dir.m.pair[0];
+    lfs_off_t off = dir.m.off;
+    lfs_dir_close(&lfs, &dir) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // tweak byte
+    uint8_t bbuffer[LFS_BLOCK_SIZE];
+    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
+
+    bbuffer[off + BYTE_OFF] = BYTE_VALUE;
+
+    cfg.erase(&cfg, block) => 0;
+    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
+
+    lfs_mount(&lfs, &cfg) => 0;
+
+    // can read?
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    // can write?
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_APPEND) => 0;
+    strcpy((char*)buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    strcpy((char*)buffer, "hello");
+    size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    strcpy((char*)buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_unmount(&lfs) => 0;
+'''