| 1 |
This directory contains tests for serialized objects from the regex-automata |
| 2 |
crate. Currently, there are only two supported such objects: dense and sparse |
| 3 |
DFAs. |
| 4 |
|
| 5 |
The idea behind these tests is to commit some serialized objects and run some |
| 6 |
basic tests by deserializing them and running searches and ensuring they are |
| 7 |
correct. We also make sure these are run under Miri, since deserialization is |
| 8 |
one of the biggest places where undefined behavior might occur in this crate |
| 9 |
(at the time of writing). |
| 10 |
|
| 11 |
The main thing we're testing is that the *current* code can still deserialize |
| 12 |
*old* objects correctly. Generally speaking, compatibility extends to semver |
| 13 |
compatible releases of this crate. Beyond that, no promises are made, although |
| 14 |
in practice callers can at least depend on errors occurring. (The serialized |
| 15 |
format always includes a version number, and incompatible changes increment |
| 16 |
that version number such that an error will occur if an unsupported version is |
| 17 |
detected.) |
| 18 |
|
| 19 |
To generate the dense DFAs, I used this command: |
| 20 |
|
| 21 |
``` |
| 22 |
$ regex-cli generate serialize dense regex \ |
| 23 |
MULTI_PATTERN_V2 \ |
| 24 |
tests/gen/dense/ \ |
| 25 |
--rustfmt \ |
| 26 |
--safe \ |
| 27 |
--starts-for-each-pattern \ |
| 28 |
--specialize-start-states \ |
| 29 |
--start-kind both \ |
| 30 |
--unicode-word-boundary \ |
| 31 |
--minimize \ |
| 32 |
'\b[a-zA-Z]+\b' \ |
| 33 |
'(?m)^\S+$' \ |
| 34 |
'(?Rm)^\S+$' |
| 35 |
``` |
| 36 |
|
| 37 |
And to generate the sparse DFAs, I used this command, which is the same as |
| 38 |
above, but with `s/dense/sparse/g`. |
| 39 |
|
| 40 |
``` |
| 41 |
$ regex-cli generate serialize sparse regex \ |
| 42 |
MULTI_PATTERN_V2 \ |
| 43 |
tests/gen/sparse/ \ |
| 44 |
--rustfmt \ |
| 45 |
--safe \ |
| 46 |
--starts-for-each-pattern \ |
| 47 |
--specialize-start-states \ |
| 48 |
--start-kind both \ |
| 49 |
--unicode-word-boundary \ |
| 50 |
--minimize \ |
| 51 |
'\b[a-zA-Z]+\b' \ |
| 52 |
'(?m)^\S+$' \ |
| 53 |
'(?Rm)^\S+$' |
| 54 |
``` |
| 55 |
|
| 56 |
The idea is to try to enable as many of the DFA's options as possible in order |
| 57 |
to test that serialization works for all of them. |
| 58 |
|
| 59 |
Arguably we should increase test coverage here, but this is a start. Note |
| 60 |
that in particular, this does not need to test that serialization and |
| 61 |
deserialization correctly roundtrips on its own. Indeed, the normal regex test |
| 62 |
suite has a test that does a serialization round trip for every test supported |
| 63 |
by DFAs. So that has very good coverage. What we're interested in testing here |
| 64 |
is our compatibility promise: do DFAs generated with an older revision of the |
| 65 |
code still deserialize correctly? |