cms authored 10 months ago and cms committed 10 months ago

no idea where this got made, but i guess i staged it by accident
i assume it's part of this very long lived CI experimentation branch
that was recently merged

Global toggle for CI jobs

See merge request !10

further experiments with type hints

- removing as much run-time checks as feasible
- the main change here is type declarations for the main `ulid` function
getting the signature right for the function argument (i.e. :random) was tricky and I don't think that it's being checked at compile time
i did learn that using 'describe' in sbcl on functions will tell you the inferred type, which is a nice head start

copied config block from CICL contained a paste error (of course)

Preparing to add release CI pipeline from CLCI

See merge request colinstrickland/cms-ulid!8

just including their definitions, adding a changlelog.md is sufficient
to set it up. Then it is triggered by tags that start with a `v*`

I'm not sure how the merging works with tags

Add run time type checks

See merge request colinstrickland/cms-ulid!7

we can construct the largest possible ULID manually, and then try and
ask for another using the same timestamp

it should signal a type error, because we have no headroom left

we're relying on the "type system" to protect us against ULID
overflows, but the compiler checks and assertions aren't guaranteed to
be portable, so we need to inspect our bounds to see that we don't
pass the upper limit for ULID according to the spec

rather than ad-hoc checks and `error` like I was using before, this
can all be simplified to use `check-type`

More unit tests, including signals tests

See merge request colinstrickland/cms-ulid!6

now i know how to make fiveam verify errrors are raised too

we can compute if the passed integer will fit in the specified byte
sequence length, and signal a type error if it does not, or if it is negative

sbcl will raise for us because of type checking but other lisps won't
necessarily so add a simple argument check that the byte we're passed
is unsigned int8

Maintenance on binary helper bits, more tests, more type hints

See merge request colinstrickland/cms-ulid!5

fiveam gives a helpful form where you can provide a predicate function
and a series of pairs for expectations and it will chain each of them.

This saves me figuring out how to write a macro to do this myself :-) !

we have a named argument in the lambda random function that we
discard, which makes a compile time warning.
declaring it as ignored for cleanliness

just a straightforward value check

sbcl will raise for us because of type checking but other lisps won't
necessarily so add a simple argument check that the byte we're passed
is unsigned int8

use the type checker to fix the input to 0 > 255
because otherwise we will silently underflow

for readability, let us be honest

it means we don't have to destructure the table arguments deeply be
can just use car and cadr

Adds a test to check int->bytes works as expected

Split the root test suite 'all-tests' into two sub suites

- 'ulid has the ulid construction tests
- 'binary-utils has tests for the binary stuff

you can run them like so

(fiveam:run! 'cms-ulid/tests::ulid)
(fiveam:run! 'cms-ulid/tests::binary-utils)
(fiveam:run! 'cms-ulid/tests::all-tests)

or , (asdf:test-system 'cms-ulid) will invoke test-op which
runs all tests

NB. while testing this iteratively on the repl, I realised that
the let over the ulid function used to initialise the `last` check
used to implement the time collision avoidance means that the static
ulid check is not idempotent - once we've run the test for the fixed
timestamp once, the ulid constructor will hold onto the fixed time

explain what they are doing, briefly

converts an 8 bit byte into individual bits

Broader tests

See merge request colinstrickland/cms-ulid!4

create a ulid from a predictable seed pair and check that it
deconstructs as expected

old habits die hard

time allows you to fix the time. random allows you to supply a random
function. Primarily useful for testing.

I want to limit the pipeline trigger to only MR

See merge request colinstrickland/cms-ulid!3

it would be nice to be able to push things to gitlab without burning minutes

Get CI working better

See merge request colinstrickland/cms-ulid!2

it's slow. At some point i could put it back for release only builds
or similar, but right now it takes ages, even though it passes fine

not doing releases yet, so I don't need those failing
CMUCL doesn't seem to work with ironclad, and I don't care about it
ABCL works but is burning minutes really

remove lisp-binary

See merge request colinstrickland/cms-ulid!1