Summary:
Previously, annotating something ThreadSafe meant "check that it is safe to run all of this procedure's methods in parallel with each other" (including self-parallelization).
This makes sense, but it means that if the user writes no annotations, we do no checking.
I'm moving toward a model of inferring when an access might happen on a thread that can run concurrently with other threads, then automatically checking that it is thread-safe w.r.t to all other accesses to the same memory (on or off the current thread thread).
This will let us report even when there are no `ThreadSafe` annotations.
Any method that is known to run on a new thread (e.g., `Runnable.run`) will be modeled as running on a thread that can run in parallel with other threads, and so will any method that is `synchronized` or acquires a lock.
In this setup, adding `ThreadSafe` to a method just means: "assume that the current method can run in parallel with any thread, including another thread that includes a different invocation of the same method (a self race) unless you see evidence to the contrary" (e.g., calling `assertMainThread` or annotating with `UiThread`).
The key step in this diff is changing the threads domain to abstract *what threads the current thread may run in parallel with* rather than *what the current thread* is. This makes things much simpler.
Reviewed By: jberdine
Differential Revision: D5895242
fbshipit-source-id: 2e23d1e
Summary:
Indicate if read or write is protected, and do not print only the
field but also the object involved in the race.
Reviewed By: sblackshear
Differential Revision: D5974250
fbshipit-source-id: 351a576
Summary:
Move Inferbo safety conditions to their own file.
Split the old `Condition.t` to a condition together with a trace.
This will ease having: different kind of condition and several traces for the same condition (see following diff)
Reviewed By: jvillard
Differential Revision: D5942030
fbshipit-source-id: d74a612
Summary:
Inject a marker using a global variable in <iostream>, and whitelist it so that
the frontend translates it.
Use the marker in the SIOF checker to tell whether a file includes <iostream>.
If so, start the analysis of its methods assuming that the standard streams are
initialised.
Reviewed By: sblackshear
Differential Revision: D5941343
fbshipit-source-id: 3388d55
Summary:
The previous domain for SIOF was duplicating some work with the generic Trace
domain, and basically was a bit confused and confusing. A sink was a set of
global accesses, and a state contains a set of sinks. Then the checker has to
needlessly jump through hoops to normalize this set of sets of accesses into a
set of accesses.
The new domain has one sink = one access, as suggested by sblackshear. This simplifies
a few things, and makes the dedup logic much easier: just grab the first report
of the list of reports for a function.
We only report on the fake procedures generated to initialise a global, and the
filtering means that we keep only one report per global.
Reviewed By: sblackshear
Differential Revision: D5932138
fbshipit-source-id: acb7285
Summary: The case of closures was not considered for the convertion of SIL instructions into HIL instructions
Reviewed By: sblackshear
Differential Revision: D5929675
fbshipit-source-id: bb6920a
Summary: The tests are slower when running in debug mode, and it creates a lot of html outputs
Reviewed By: sblackshear
Differential Revision: D5916511
fbshipit-source-id: 07c90b7
Summary:
This diff does two things:
# Infer no longer add the contrains that the return value of a skip function is never null. This was leading to false negatives and is not necessary as those return value are treated angelically
# Infer now support `Nonnull` on the return value of skip functions.
Reviewed By: jberdine, sblackshear
Differential Revision: D5840324
fbshipit-source-id: bbd8d82
Summary: Example of combination between annotating fields with nullable and the biabduction analysis in Objective C
Reviewed By: dulmarod
Differential Revision: D5906016
fbshipit-source-id: b95c6e0
Summary:
The interval bound of the abstract domain is extended.
`[min|max](int, symbol)` => `int [+|-] [min|max](int, symbol)`
As a result, `vector::empty` can be modelled to return a more precise value: `1 - min(1, size)`.
Reviewed By: mbouaziz
Differential Revision: D5899404
fbshipit-source-id: c8c3f49
Summary:
We take it into account to not report bugs inside the available block. This requires a plugin change.
update-submodule: facebook-clang-plugins
Reviewed By: jvillard
Differential Revision: D5891511
fbshipit-source-id: 21a02ad
Summary:
With this change and the previous facebook-clang-plugins change, infer no
longer exhausts the biniou buffer when reading the serialized C++ AST.
update-submodule: facebook-clang-plugins
Reviewed By: mbouaziz
Differential Revision: D5891081
fbshipit-source-id: cf48eac
Summary: Infer can then detect the resource leak when resources are stored into a container
Reviewed By: sblackshear
Differential Revision: D5887702
fbshipit-source-id: 6106cfb
Summary: The point of the tracing mode is to compute all the possible path leading to an error state. However, within a method, many of those paths are not feasibile in practice. This leads to many false alarms for the resource leak analysis.
Reviewed By: sblackshear
Differential Revision: D5888695
fbshipit-source-id: 2dbc57b
Summary: Model the `remove(...)` and the `clear()` method of `HashMap`.
Reviewed By: sblackshear
Differential Revision: D5887674
fbshipit-source-id: c3f40ee
Summary: Handling the utility functions for asserting that we're on background thread.
Reviewed By: jberdine
Differential Revision: D5863435
fbshipit-source-id: 3ad95b5
Summary:
Previously, we just tracked a boolean representing whether we were possibly on the main thread (true) or definitely not on the main thread (false).
In order to start supporting `Thread.start`, `Runnable.run`, etc., we'll need something more expressive.
This diff introduces a lattice:
```
Any
/ \
Main Background
\ /
Unknown
```
as the new threads domain. The initial value is `Unknown`, and we introduce `Main` in situations where we would have introduced `true` before.
This (mostly) preserves behavior: the main difference is that before code like
```
if (*) {
assertMainThread()
} else {
x.f = ...
}
```
would have recorded that the access to `x.f` was on the main thread, whereas now we'll say that it's on an unknown thread.
Reviewed By: peterogithub
Differential Revision: D5860256
fbshipit-source-id: efee330
Summary:
It's useful to be able to disable de-duplication on the command line with `--no-filtering`.
Gate de-duplication with `Config.filtering` and move the de-duplication tests to a new directory under the build systems tests.
Reviewed By: jeremydubreil
Differential Revision: D5865329
fbshipit-source-id: 5094f5b
Summary:
Since D5381239, infer is careful not to delete directories that do not "look
like" results directories on startup, in case the user passed, eg, `-o /`.
In our repo, lots of results dir are created by build/test of infer, and when
the version of infer changes and the expected contents of results directories
change then it might start refusing to delete the results directories created
with another version of infer.
Add an option to force infer to delete the results directory no matter how
dodgy it looks, and use it in our repo by adding the option in every
.inferconfig.
Reviewed By: mbouaziz
Differential Revision: D5870984
fbshipit-source-id: 09412de
Summary: Will then be easier to understand if some changes in the test results is legit or not.
Reviewed By: sblackshear
Differential Revision: D5863961
fbshipit-source-id: 7eb3f33
Summary:
Suggesting to add `_Nullable` on the fields checked for, or assigned to, `nullptr` will allow the biabduction analysis to report null dereferences that are related to the lifetime of objects.
Depends on D5832147
Reviewed By: sblackshear
Differential Revision: D5836538
fbshipit-source-id: c1b8e48
Summary: The prune nodes where translated as `prune (expr = false)` and `prune ( expr != false)`. This case is a bit tricky to deconstruct in HIL. This diff translates the prune instructions as just `prune !expr` for the true branch and `prune expr` for the false branch.
Reviewed By: dulmarod
Differential Revision: D5832147
fbshipit-source-id: 2c3502d
Summary:
We need to make sure that destructors of virtual base classes are called only once. Similarly to what clang does, we have two destructors for a class: a destructor wrapper and an inner destructor.
Destructor wrapper is called from outside, i.e., when variables go out of scope or when destructors of fields are being called.
Destructor wrappers have calls to inner destructors of all virtual base classes in the inheritance as their bodies.
Inner destructors have destructor bodies and calls to destructor wrappers of fields and inner destructors of non-virtual base classes.
Reviewed By: dulmarod
Differential Revision: D5834555
fbshipit-source-id: 51db238
Summary:
The "placement new" operator `new (e) T` constructs a `T` in the pre-allocated memory address `e`.
We weren't translating the `e` part, which was leading to false positives in the dead store analysis.
Reviewed By: dulmarod
Differential Revision: D5814191
fbshipit-source-id: 05c6fa9
Summary:
Simple instance of the problem: analyzing the following program times out.
```
#include <tuple>
void foo() {
std::tuple<std::tuple<int>> x;
}
```
Replacing `std::tuple<std::tuple<int>>` by `std::tuple<int>` makes the analysis
terminate.
In the AST, both tuple<tuple<int>> and tuple<int> have the same template
specialization type: "Pack" (which means we're supposed to go look into the
arguments of the template to get their values). This is not information enough
and that's the plugin fault.
On the backend side, this means that two types have the same Typ.Name.t, namely
"std::tuple<_>", so they collide in the tenv. The definition of
tuple<tuple<int>> is the one making it into the tenv. One of the fields of the
corresponding CxxRecord is of type "tuple<int>", which we see as the same
"tuple<_>", which causes the loop.
update-submodule: facebook-clang-plugins
Reviewed By: mbouaziz
Differential Revision: D5775840
fbshipit-source-id: 0528604
Summary:
Sort the complete set of warnings by everything except procname, then de-duplicate.
This scheme prevents reporting identical error messages on the same line/same file.
This is important for avoiding duplicate reports on multiple instantiations of the same template.
Reviewed By: jberdine
Differential Revision: D5819467
fbshipit-source-id: 984f47f
Summary: The resolution was previously only happening for constructors, but calls to private methods or to `super` are also neither static calls nor virtual calls. In this case, the resolution logic should be the same as for constructors.
Reviewed By: sblackshear
Differential Revision: D5830376
fbshipit-source-id: 9b56f80
Summary: Destroying local variables that are out of scope after `continue`.
Reviewed By: jberdine
Differential Revision: D5804120
fbshipit-source-id: 638cff5
Summary: Destroying local variables that are out of scope after `break`.
Reviewed By: jberdine
Differential Revision: D5764647
fbshipit-source-id: a7e06ae
Summary: The successor node of `continue` was not correct inside the `do while`.
Reviewed By: sblackshear
Differential Revision: D5769578
fbshipit-source-id: d7b0843
Summary: With this diff, the analysis trace will jump to the definition of the skipped methods when the location is known. This is especially useful when the analysis is relying on the method annotations.
Reviewed By: sblackshear
Differential Revision: D5783428
fbshipit-source-id: 561b739
Summary:
We supported globals as sources before, but we did so by allowing ClangTrace etc. to match against any access path in the footprint of the trace.
This is very powerful/flexible, but it's ultimately not a good idea because it leads to traces that are hard to read.
This is because a footprint source doesn't have any information about its provenance: we might know that the value came from a global, but we don't know where the read occurred.
The mechanism for handling procedure calls as sources already knows how to solve this problem.
This diff implements globals as sources as a special case of procedure call sources instead.
This will give us much nicer traces with full provenance of the read from the global.
Reviewed By: mbouaziz
Differential Revision: D5772299
fbshipit-source-id: 491ae81
Summary: The list of fields of a Java object in SIL is the list of fields declared in the class plus all the fields declared in the the super classes. It turns out that we were missing the fields declared in the implemented interfaces.
Reviewed By: sblackshear
Differential Revision: D5720386
fbshipit-source-id: d65c9de
Summary:
When a lambda has an `auto` parameter, the inferred type of the parameter because part of the name.
Our heuristic for identifying lambda was checking if the lambda's name was exactly `operator()`, which won't catch this case.
Reviewed By: jvillard
Differential Revision: D5753323
fbshipit-source-id: 85ff75a
Summary:
Not translating these properly was causing false positives for the dead store analysis in cases like
```
int i = 0;
return [j = i]() { return j; }();
```
Reviewed By: da319
Differential Revision: D5731562
fbshipit-source-id: ae79ac8
Summary: We inject destructor calls of base classes inside destructor bodies after the destructor calls of fields.
Reviewed By: dulmarod
Differential Revision: D5745499
fbshipit-source-id: 90745ec
Summary: We used to crash whenever we hit these. The simple translation implemented here is not particularly inspiring, but it is better than crashing.
Reviewed By: jvillard
Differential Revision: D5702095
fbshipit-source-id: 3795d43
Summary: With this, we can now get now get inter-procedural issues involving native methods.
Reviewed By: sblackshear
Differential Revision: D5730638
fbshipit-source-id: 3bdbdbd
Summary: Atoms of the form `identifier = footprint var` naturally occurs with the angelic analysis mode. So it is not clear to me why we should drop those.
Reviewed By: sblackshear
Differential Revision: D5654754
fbshipit-source-id: 9dd2eb5
Summary: This makes the traces more readable when involving skipped functions.
Reviewed By: sblackshear
Differential Revision: D5731683
fbshipit-source-id: 49d363b
Summary:
A function can both be a sink and propagate source info, but we currently ignore the summary for any function that is also a sink.
This will cause us to under-report for (e.g.) `src1 = source(); src2 = strcpy(dest, src1); exec(src2)`.
This is both a potential buffer overflow and a potential shell injection, but we won't report the second issue.
Reviewed By: jberdine
Differential Revision: D5676167
fbshipit-source-id: 232ab2f
Summary:
In looking at summaries that Quandary took a long time to compute, one thing I notice frequently is redundancy in the footprint sources (e.g., I might see `Footprint(x), Footprint(x.f), Footprint(x*)`).
`sudo perf top` indicates that joining big sets of sources is a major performance bottleneck, and a large number of footprint sources is surely a big part of this (since we expect the number of non-footprint sources to be small).
This diff addresses the redundancy issue by using a more complex representation for a set of sources. The "known" sources are still in a set, but the footprint sources are now represented as a set of access paths (via an access trie).
The access path trie is a minimal representation of a set of access paths, so it would represent the example above as a simple `x*`.
This should make join/widen/<= faster and improve performance
Reviewed By: jberdine
Differential Revision: D5663980
fbshipit-source-id: 9fb66f8
Summary:
The previous widening operator added stars to the *end* of paths that existed in `next` but not `prev`. This is not enough to ensure termination in the case where the trie is growing both deeper and wider at the same time.
The newly added test demonstrates this issue. In the code, there's an ever-growing path of the form `tmp.prev.next.prev.next...` that wasn't summarized by the previous widening operator. The new widening is much more aggressive: it replaces *any* node present in `next` but not `prev` with a `*` (rather than trying to tack a star onto the end). This fixes the issue.
This issue was causing divergence on tricky doubly-linked list code in prod.
Reviewed By: jeremydubreil
Differential Revision: D5665719
fbshipit-source-id: 1310a92
Summary:
This is a check for when an unavailable class is being allocated.
This diff also adds a check for the context to remove false positives: If the class is not available but the method calls are wrapped in a check whether the class is available, then don't report.
Reviewed By: jvillard
Differential Revision: D5631191
fbshipit-source-id: 2082dfe
Summary:
Calling exit at the end of a proc does not create unreachable code, prior to this commit inferbo reports that it does.
We extend collect_instrs to detect when we're at the end of a procedure in C and not report on unreachable code if we call a procedure there.
Reviewed By: mbouaziz
Differential Revision: D5623637
fbshipit-source-id: 0d5f326
Summary: This check is not possible in Java as it natirally happens in the totally legit case of the `try ... finally`.
Reviewed By: sblackshear
Differential Revision: D5568802
fbshipit-source-id: 24ca074
Summary:
Instead of a whitelist and blacklist and default issue types and default
blacklist and filtering, consider a simpler semantics where
1. checkers can be individually turned on or off on the command line
2. most checkers are on by default
3. `--no-filtering` turns all issue types on, but they can then be turned off again by further arguments
This provides a more flexible CLI and is similar to other options in the infer
CLI, where "global" behaviour is generally avoided.
Dynamically created checkers (eg, AL linters) cause some complications in the
implementation but I think the semantics is still clear.
Also change the name of the option to mention "issue types" instead of
"checks", since the latter can be confused with "checkers".
Reviewed By: jberdine
Differential Revision: D5583238
fbshipit-source-id: 21de476
Summary:
We previously lumped ownership predicates in with all other predicates. That limited us to a flat ownership domain.
This diff separates out the ownership predicates so we can have a richer lattice of predicates with each access path.
This lets us be more precise; for example, we can now show that
```
needToOwnBothParams(Obj o1, Obj o2) {
Obj alias;
if (*) { alias = o1; } else { alias = o2; }
alias.f = ... // both o1 and o2 need to be owned for this to be safe
}
void ownBothParamsOk() {
needToOwnBothParams(new Obj(), new Obj()); // ok, would have complained before
}
```
is safe.
Reviewed By: jberdine
Differential Revision: D5589898
fbshipit-source-id: 9606a46
Summary:
This makes it easier to test a single checker.
Also refactor the code to make it harder to mess up the list of default/all checkers.
Reviewed By: sblackshear
Differential Revision: D5583209
fbshipit-source-id: 7c919b2
Summary:
Use jbuilder to build infer instead of ocamlbuild. This is mainly to get faster builds:
```
times in 10ms, ±differences measured in speedups, 4 cores
| | ocb total | jb | ±total | ocb user | jb | ±user | ocb cpu | jb | ±cpu | ocb sys | jb | ±sys |
|-----------------------------------+-----------+------+--------+----------+------+-------+---------+-----+------+---------+------+------|
| byte from scratch | 6428 | 2456 | 2.62 | 7743 | 6662 | 1.16 | 138 | 331 | 2.40 | 1184 | 1477 | 0.80 |
| native from scratch | 9841 | 4289 | 2.29 | 9530 | 8834 | 1.08 | 110 | 245 | 2.23 | 1373 | 1712 | 0.80 |
| byte after native | 29578 | 1602 | 18.46 | 4514 | 4640 | 0.97 | 170 | 325 | 1.91 | 543 | 576 | 0.94 |
| change infer.ml byte | 344 | 282 | 1.22 | 292 | 215 | 1.36 | 96 | 99 | 1.03 | 040 | 066 | 0.61 |
| change infer.ml native | 837 | 223 | 3.75 | 789 | 174 | 4.53 | 98 | 99 | 1.01 | 036 | 47 | 0.77 |
| change Config.ml byte | 451 | 339 | 1.33 | 382 | 336 | 1.14 | 97 | 122 | 1.26 | 056 | 80 | 0.70 |
| change Config.ml native | 4024 | 1760 | 2.29 | 4585 | 4225 | 1.09 | 127 | 276 | 2.17 | 559 | 644 | 0.87 |
| change cFrontend_config.ml byte | 348 | 643 | 0.54 | 297 | 330 | 0.90 | 96 | 67 | 0.70 | 038 | 102 | 0.37 |
| change cFrontend_config.ml native | 1480 | 584 | 2.53 | 1435 | 906 | 1.58 | 106 | 185 | 1.75 | 136 | 178 | 0.76 |
#+TBLFM: $4=$2/$3;f2::$7=$5/$6;f2::$10=$9/$8;f2::$13=$11/$12;f2
50 cores
| | ocb total | jb | ±total | ocb user | jb | ±user | ocb cpu | jb | ±cpu | ocb sys | jb | ±sys |
|---------------------+-----------+------+--------+----------+------+-------+---------+----+------+---------+------+------|
| byte from scratch | 9114 | 2061 | 4.42 | 9334 | 5133 | 1.82 | | | 0/0 | 2566 | 1726 | 1.49 |
| native from scratch | 13481 | 3967 | 3.40 | 12291 | 7608 | 1.62 | | | 0/0 | 3003 | 2100 | 1.43 |
| byte after native | 3467 | 1476 | 2.35 | 5067 | 3912 | 1.30 | | | 0/0 | 971 | 801 | 1.21 |
#+TBLFM: $4=$2/$3;f2::$7=$5/$6;f2::$10=$9/$8;f2::$13=$11/$12;f2
```
Menu:
1. Write a jbuild file, autogenerated from jbuild.in because we need to fill in
some information at build-time (really, at configure time, but TODO), such as
whether or not clang is enabled.
2. Nuke lots of stuff from infer/src/Makefile that is now in the jbuild file
3. The jbuild file lives in infer/src/ so it can see all the sources. If we put it somewhere else, eg, infer/, then `jbuilder` scans too many files (all irrelevant) and takes 2.5s to start instead of .8s. Adding irrelevant directories to jbuild-ignore does not help.
4. jbuilder does not support subdirectories, so resort to listing all the
source files in the generated jbuild (only source directories need to be
manually listed in jbuild.in though). Still, the generated .merlin is wrong
and makes merlin find source files in _build, so manually tune it to get
good merlin support. We also lose some of merlin for unit tests as it
cannot see their build artefacts anymore.
5. checkCopyright gets its own jbuild because it's standalone. Also, remove
some deprecation warnings in checkCopyright due to the new version of Core from
a while ago.
6. Drop less-used Makefile features (they had regressed anyway) such as
building individual modules. Also, building mod_dep.pdf now takes all the
source files available so they better build (before, it would only take the
source files from the config, eg with or without clang) (that's pretty minor).
7. The toplevel is now built as a custom toplevel because that was easier. It
should soon be even easier: https://github.com/janestreet/jbuilder/issues/210
8. Move BUILTINS.mli to BUILTINS.ml because jbuilder is not happy about
interface files without implementations.
In particular, I did not try to migrate too much of the Makefile logic to jbuilder,
more can be done in the future.
Reviewed By: jberdine
Differential Revision: D5573661
fbshipit-source-id: 4ca6d8f
Mehdi Bouaziz
Sam Blackshear
Josh Berdine
Jules Villard
Dulma Churchill
Jeremy Dubreil
Sungkeun Cho
Dino Distefano
Daiva Naudziuniene
Roy Li
Greg Nisbet
Jia Chen
Meng Zhang