Prevent setting end time in the past

[andreivladbrg]

Addresses https://github.com/cantinasec/sablier/issues/7

[PaulRBerg]

Feedback below.

[andreivladbrg]

I have improved the wording in my last commit. Can you please confirm that everything is good now?

[PaulRBerg]

I approved the changes because I have already resolved the comments with my latest commits.

But please have a look at all comments, @andreivladbrg.

the warp cheatcode doesn't work for the invariant tests

You are right. What we had to was to bound the start time and the cliff time like this:

uint40 currentTime = getBlockTimestamp();
params.broker.fee = bound(params.broker.fee, 0, DEFAULT_MAX_FEE);
params.range.start = boundUint40(params.range.start, 0, currentTime);
params.range.cliff = boundUint40(params.range.cliff, currentTime, currentTime + 52 weeks);
params.range.end = boundUint40(params.range.end, params.range.cliff + 1, MAX_UNIX_TIMESTAMP);

Another pre-requisite to make this work was to increase the default block.timestamp used in our tests. I ended up setting it to March 1, 2023:

DEFAULT_START_TIME = uint40(1_677_632_400);
vm.warp({ timestamp: DEFAULT_START_TIME });

This proved to be a very good decision, since it has led me to identify several minor issues in the tests (e.g. using DEFAULT_CLIFF_TIME instead of DEFAULT_CLIFF_DURATION in the bounding of timeWarp).

In addition, setting the block timestamp to a value in the present creates a more realistic testing environment (as I have argued here).

[andreivladbrg]

DEFAULT_CLIFF_TIME instead of DEFAULT_CLIFF_DURATION

Good catch

setting the block timestamp to a value in the present creates a more realistic testing environment

TBH, I don't have an opinion here . Feel free to proceed as you see want.

[PaulRBerg]

There is an invariant test run that failed, but it is unrelated to this PR (see my explanation here).