Test suites already had the ability to provide suite-level code via the
"code" attribute, but this was placed in the suite's generated source
file, making it inaccessbile to internal tests.
This change allows suite code to be placed in the same place as internal
tests, via the "in" attribute, though this has some caveats:
1. Suite-level code generally declares helper functions in global scope.
We don't parse this code or anything, so name collisions between
helper functions across different test suites is up to the developer
to resolve.
2. Internal suite-level code has access to internal functions/variables/
etc, this means we can't place a copy in our suite's generate source
and expect it to compile. For this reason, internal suite-level code
is unavailable for non-internal tests in the suite.
This also means you only get to place internal suite-level code in a
single source file. Though this is not really an issue since littlefs
is basically a single file...
The previous system of relying on test name prefixes for ordering was
simple, but organizing tests by dependencies and topologically sorting
during compilation is 1. more flexible and 2. simplifies test names,
which get typed a lot.
Note these are not "hard" dependencies, each test suite should work fine
in isolation. These "after" dependencies just hint an ordering when all
tests are ran.
As such, it's worth noting the tests should NOT error of a dependency is
missing. This unfortunately makes it a bit hard to catch typos, but
allows faster compilation of a subset of tests.
---
To make this work the way tests are linked has changed from using custom
linker section (fun linker magic!) to a weakly linked array appended to
every source file (also fun linker magic!).
At least with this method test.py has strict control over the test
ordering, and doesn't depend on 1. the order in which the linker merges
sections, and 2. the order tests are passed to test.py. I didn't realize
the previous system was so fragile.
Instead of iterating over a number of seeds in the test itself, the
seeds are now permuted as a part of normal test defines.
This lets each seed take advantage of other test features, mainly the
ability to test powerlosses heuristically.
This is probably how it should have been done in the first place, but
the permutation tests can't do this since the number of permutations
changes as the size of the test input changes. The test define system
can't handle that very well.
The tradeoffs here are:
- We can't do cross-fuzz checks, such as the balance checks in the rbyd
tests, though those really should be moved to benchmarks anyways.
- The large number of cheap fuzz permutations skews the total
permutation count, though I'm not sure this matters.
before: 3083 permutations (-Gnor)
after: 409893 permutations (-Gnor)
Any conditions in both the suites and cases are anded together to
determine when the test/bench should run.
Accepting a list here makes it easier to compose multiple conditions,
since toml-level elements are a bit easier to modify than strings of
C expressions.
This marks internal tests/benches (case.in="lfs.c") with an otherwise-unused
flag that is printed during --summary/--list-*. This just helps identify which
tests/benches are internal.
Previously no matches would noop, which, while consistent with an empty
test suite that contains no tests but shouldn't really error, this made
it easy to miss when a typo would cause tests to be missed.
Also added a bit of color to script-level errors in test/bench.py
This reworks test.py/bench.py a bit to map arguments to ids as a first
step instead of defering as much as possible. This is a better design
and avoids the hackiness around -b/-B. As a plus, test_id globbing is
easy to add.
I wondered if walking in Python 2's footsteps was going to run into the
same issues and sure enough, memory backed iterators became unweildy.
The motivation for this change is that large ranges in tests, such as
iterators over seeds or permutations, became prohibitively expensive to
compile. This meant more iteration moving into tests with more steps to
reproduce failures. This sort of defeats the purpuse of the test
framework.
The solution here is to move test permutation generation out of test.py
and into the test runner itself. The allows defines to generate their
values programmatically.
This does conflict with the test frameworks support of sets of explicit
permutations, but this is fixed by also moving these "permutation sets"
down into the test runner.
I guess it turns out the closer your representation matches your
implementation the better everythign works.
Additionally the define caching layer got a bit of tweaking. We can't
precalculate the defines because of mutual recursion, but we can
precalculate which define/permutation each define id maps to. This is
necessary as otherwise figuring out each define's define-specific
permutation would be prohibitively expensive.
- Added both uattr (limited to 256) and id (limited to 65535) benchmarks
covering the main rbyd operations
- Fixed issue where --defines gets passed to the test/bench runners when
querying id-specific information. After changing the test/bench
runners to prioritize explicit defines, this causes problems for
recorded benchmark results and debug related things.
- In plot.py/plotmpl.py, made --by/-x/-y in subplots behave somewhat
reasonably, contributing to a global dataset and the figure's legend,
colors, etc, but only shown in the specified subplot. This is useful
mainly for showing different -y values on different subplots.
- In plot.py/plotmpl.py, added --labels to allow explicit configuration
of legend labels, much like --colors/--formats/--chars/etc. This
removes one of the main annoying needs for modifying benchmark results.
- Added support for negative numbers in the leb16 encoding with an
optional 'w' prefix.
- Changed prettyasserts.py rule to .a.c => .c, allowing other .a.c files
in the future.
- Updated .gitignore with missing generated files (tags, .csv).
- Removed suite-namespacing of test symbols, these are no longer needed.
- Changed test define overrides to have higher priority than explicit
defines encoded in test ids. So:
./runners/bench_runner bench_dir_open:0f1g12gg2b8c8dgg4e0 -DREAD_SIZE=16
Behaves as expected.
Otherwise it's not easy to experiment with known failing test cases.
- Fixed issue where the -b flag ignored explicit test/bench ids.
For long running processes (testing with >1pls) these logs can grow into
multiple gigabytes, humorously we never access more than the last n lines
as requested by --context. Piping the stdout with --stdout does not use
additional RAM.
This happens in rare situations where there is a failed mdir relocation,
interrupted by a power-loss, containing the destination of a directory
rename operation, where the directory being renamed preceded the
relocating mdir in the mdir tail-list. This requires at some point for a
previous directory rename to create a cycle.
If this happens, it's possible for the half-orphan to contain the only
reference to the renamed directory. Since half-orphans contain outdated
state when viewed through the mdir tail-list, the renamed directory
appears to be a full-orphan until we fix the relocating half-orphan.
This causes littlefs to incorrectly remove the renamed directory from
the mdir tail-list, causes catastrophic problems down the line.
The source of the problem is that the two different types of orphans
really operate on two different levels of abstraction: half-orphans fix
failed mdir commits, while full-orphans fix directory removes/renames.
Conflating the two leads to situations where we attempt to fix assumed
problems about the directory tree before we have fixed problems with the
mdir state.
The fix here is to separate out the deorphan search into two passes: one
to fix half-orphans and correct any mdir-commits, restoring the mdirs
and gstate to a known good state, then two to fix failed
removes/renames.
---
This was found with the -Plinear heuristic powerloss testing, which now
runs on more geometries. The failing case was:
test_relocations_reentrant_renames:112gg261dk1e3f3:123456789abcdefg1h1i1j1k1
l1m1n1o1p1q1r1s1t1u1v1g2h2i2j2k2l2m2n2o2p2q2r2s2t2
Also fixed/tweaked some parts of the test framework as a part of finding
this bug:
- Fixed off-by-one in exhaustive powerloss state encoding.
- Added --gdb-powerloss-before and --gdb-powerloss-after to help debug
state changes through a failing powerloss, maybe this should be
expanded to any arbitrary powerloss number in the future.
- Added lfs_emubd_crc and lfs_emubd_bdcrc to get block/bd crcs for quick
state comparisons while debugging.
- Fixed bd read/prog/erase counts not being copied during exhaustive
powerloss testing.
- Fixed small typo in lfs_emubd trace.
- Changed --(tool)-tool to --(tool)-path in scripts, this seems to be
a more common name for this sort of flag.
- Changed BUILDDIR to not have implicit slash, makes Makefile internals
a bit more readable.
- Fixed some outdated names hidden in less-often used ifdefs.
- Fixed prettyasserts.py parsing when '->' is in expr
- Made prettyasserts.py failures not crash (yay dynamic typing)
- Fixed the initial state of the emubd disk file to match the internal
state in RAM
- Fixed true/false getting changed to True/False in test.py/bench.py
defines
- Fixed accidental substring matching in plot.py's --by comparison
- Fixed a missed LFS_BLOCk_CYCLES in test_superblocks.toml that was
missed
- Changed test.py/bench.py -v to only show commands being run
Including the test output is still possible with test.py -v -O-, making
the implicit inclusion redundant and noisy.
- Added license comments to bench_runner/test_runner
Based loosely on Linux's perf tool, perfbd.py uses trace output with
backtraces to aggregate and show the block device usage of all functions
in a program, propagating block devices operation cost up the backtrace
for each operation.
This combined with --trace-period and --trace-freq for
sampling/filtering trace events allow the bench-runner to very
efficiently record the general cost of block device operations with very
little overhead.
Adopted this as the default side-effect of make bench, replacing
cycle-based performance measurements which are less important for
littlefs.
This provides 2 things:
1. perf integration with the bench/test runners - This is a bit tricky
with perf as it doesn't have its own way to combine perf measurements
across multiple processes. perf.py works around this by writing
everything to a zip file, using flock to synchronize. As a plus, free
compression!
2. Parsing and presentation of perf results in a format consistent with
the other CSV-based tools. This actually ran into a surprising number of
issues:
- We need to process raw events to get the information we want, this
ends up being a lot of data (~16MiB at 100Hz uncompressed), so we
paralellize the parsing of each decompressed perf file.
- perf reports raw addresses post-ASLR. It does provide sym+off which
is very useful, but to find the source of static functions we need to
reverse the ASLR by finding the delta the produces the best
symbol<->addr matches.
- This isn't related to perf, but decoding dwarf line-numbers is
really complicated. You basically need to write a tiny VM.
This also turns on perf measurement by default for the bench-runner, but at a
low frequency (100 Hz). This can be decreased or removed in the future
if it causes any slowdown.
- Changed multi-field flags to action=append instead of comma-separated.
- Dropped short-names for geometries/powerlosses
- Renamed -Pexponential -> -Plog
- Allowed omitting the 0 for -W0/-H0/-n0 and made -j0 consistent
- Better handling of --xlim/--ylim
These are really just different flavors of test.py and test_runner.c
without support for power-loss testing, but with support for measuring
the cumulative number of bytes read, programmed, and erased.
Note that the existing define parameterization should work perfectly
fine for running benchmarks across various dimensions:
./scripts/bench.py \
runners/bench_runner \
bench_file_read \
-gnor \
-DSIZE='range(0,131072,1024)'
Also added a couple basic benchmarks as a starting point.
- Added the littlefs license note to the scripts.
- Adopted parse_intermixed_args everywhere for more consistent arg
handling.
- Removed argparse's implicit help text formatting as it does not
work with perse_intermixed_args and breaks sometimes.
- Used string concatenation for argparse everywhere, uses backslashed
line continuations only works with argparse because it strips
redundant whitespace.
- Consistent argparse formatting.
- Consistent openio mode handling.
- Consistent color argument handling.
- Adopted functools.lru_cache in tracebd.py.
- Moved unicode printing behind --subscripts in traceby.py, making all
scripts ascii by default.
- Renamed pretty_asserts.py -> prettyasserts.py.
- Renamed struct.py -> struct_.py, the original name conflicts with
Python's built in struct module in horrible ways.
The main benefit is small test ids everywhere, though this is with the
downside of needing longer names to properly prefix and avoid
collisions. But this fits into the rest of the scripts with globally
unique names a bit better. This is a C project after all.
The other small benefit is test generators may have an easier time since
per-case symbols can expect to be unique.
This is really more work for the bench runner. With this change defines
can be manipulated at a rather high level at runtime. Which should be
useful for generating benchmarks across various dimensions.
The define grammar in the test_runner is now a bit more powerful,
accepting:
1. A single value: -DN=42
2. A list of values, which get permuted: -DN=1,2,3
3. A range: -DN=range(10)
4. Some combo: -DN=1,2,range(3,0,-1)
This is more complex in the test .toml defines, which can also be C
expressions:
1. A single value: define=42
2. A single expression: define='42*42'
3. A list: define=[1,2,3]
4. A comma separated string: define='1,2,3'
5. A range: define='42*range(10)'
6. This mess: define=[1,2,'3,4,range(2)*range(2)+3']
This is probably how the test runner should have been implemented in the
first place, but it took a few tries to get here.
This makes it so the test identifier, which is a bit longer now, fully
encodes the state of the defines in the test. This removes the need for
the extra geometry field and allows reproduction of tests with custom
defines at runtime.
The test runner may have already seemed like a solved problem, but these
changes are really to enable repurposing the test runner as a bench
runner.
These are just some minor quality of life improvements
- Added a "make build-test" alias
- Made test runner a positional arg for test.py since it is almost
always required. This shortens the command line invocation most of the
time.
- Added --context to test.py
- Renamed --output in test.py to --stdout, note this still merges
stderr. Maybe at some point these should be split, but it's not really
worth it for now.
- Reworked the test_id parsing code a bit.
- Changed the test runner --step to take a range such as -s0,12,2
- Changed tracebd.py --block and --off to take ranges
Previously didn't think this would work without making test.py aware of
the number of implicit defines, which risks being incredibly fragile.
Fortunately it turns out we can defer the actual array size calculation
until the C preprocessor. This simplifies a few things.
Also a bitmap-based caching layer for the defines. Since the test
defines have been upgraded to callbacks recursive defines risk spending
a decent amount of time evaluating on every lookup. Some quick testing
shows 408015154 hits to 46160 misses so that's a good sign.
Also changed the geometries to be their own leb16-encoded part of the
test identifier. This means any geometry can be captured and reproduced
with just the test identifier. Here are the current test geometries:
./runners/test_runner --list-geometries
geometry read prog erase count size leb16
d,default 16 16 512 2048 1048576 g1gg2
e,eeprom 1 1 512 2048 1048576 1gg2
E,emmc 512 512 512 2048 1048576 gg2
n,nor 1 1 4096 256 1048576 1ggg1
N,nand 4096 4096 32768 32 1048576 ggg1ggg8
Based on a handful of local hacky variations, this sort of trace
rendering is surprisingly useful for getting an understanding of how
different filesystem operations interact with the underlying
block-device.
At some point it would probably be good to reimplement this in a
compiled language. Parsing and tracking the trace output quickly
becomes a bottleneck with the amount of trace output the tests
generate.
Note also that since tracebd.py run on trace output, it can also be
used to debug logged block-device operations post-run.
This mostly involved futzing around with some of the less intuitive
parts of Unix's named-pipes behavior.
This is a bit important since the tests can quickly generate several
gigabytes of trace output.
These have no real purpose other than slowing down the simulation
for inspection/fun.
Note this did reveal an issue in pretty_asserts.py which was clobbering
feature macros. Added explicit, and maybe a bit hacky, #undef _FEATURE_H
to avoid this.
With more features being added to test.py, the one-line status is
starting to get quite long and pass the ~80 column readability
heuristic. To make this worse this clobbers the terminal output
when the terminal is not wide enough.
Simple solution is to disable line-wrapping, potentially printing
some garbage if line-wrapping-disable is not supported, but also
printing a final status update to fix any garbage and avoid a race
condition where the script would show a non-final status.
Also added --color which disables any of this attempting-to-be-clever
stuff.
The main change here from the previous test framework design is:
1. Powerloss testing remains in-process, speeding up testing.
2. The state of a test, included all powerlosses, is encoded in the
test id + leb16 encoded powerloss string. This means exhaustive
testing can be run in CI, but then easily reproduced locally with
full debugger support.
For example:
./scripts/test.py test_dirs#reentrant_many_dir#10#1248g1g2 --gdb
Will run the test test_dir, case reentrant_many_dir, permutation #10,
with powerlosses at 1, 2, 4, 8, 16, and 32 cycles. Dropping into gdb
if an assert fails.
The changes to the block-device are a work-in-progress for a
lazily-allocated/copy-on-write block device that I'm hoping will keep
exhaustive testing relatively low-cost.
This also adds coverage support to the new test framework, which due to
reduction in scope, no longer needs aggregation and can be much
simpler. Really all we need to do is pass --coverage to GCC, which
builds its .gcda files during testing in a multi-process-safe manner.
The addition of branch coverage leverages information that was available
in both lcov and gcov.
This was made easier with the addition of the --json-format to gcov
in GCC 9.0, however the lax backwards compatibility for gcov's
intermediary options is a bit concerning. Hopefully --json-format
sticks around for a while.
A small mistake in test.py's control flow meant the failing test job
would succesfully kill all other test jobs, but then humorously start
up a new process to continue testing.
This simplifies the interaction between code generation and the
test-runner.
In theory it also reduces compilation dependencies, but internal tests
make this difficult.
This mostly required names for each test case, declarations of
previously-implicit variables since the new test framework is more
conservative with what it declares (the small extra effort to add
declarations is well worth the simplicity and improved readability),
and tweaks to work with not-really-constant defines.
Also renamed test_ -> test, replacing the old ./scripts/test.py,
unfortunately git seems to have had a hard time with this.
As noted in Python's subprocess library:
> This will deadlock when using stdout=PIPE and/or stderr=PIPE and the
> child process generates enough output to a pipe such that it blocks
> waiting for the OS pipe buffer to accept more data.
Curiously, this only became a problem when updating to Ubuntu 20.04
in CI (python3.6 -> python3.8).
- Added -L/--depth argument to show dependencies for scripts/stack.py,
this replaces calls.py
- Additional internal restructuring to avoid repeated code
- Removed incorrect diff percentage when there is no actual size
- Consistent percentage rendering in test.py
I confirmed that the same number of tests are run
with "make test" on:
* Ubuntu with and without this change
* macOS with this change
> ====== results ======
> tests passed 817/817 (100.00%)
> tests failed 0/817 (0.00%)
Currently this is just lfs.c and lfs_util.c. Previously this included
the block devices, but this meant all of the scripts needed to
explicitly deselect the block devices to avoid reporting build
size/coverage info on them.
Note that test.py still explicitly adds the block devices for compiling
tests, which is their main purpose. Humorously this means the block
devices will probably be compiled into most builds in this repo anyways.
