L-09 Rounding Error Might Prevent Borrowing at the Maximum Limit

[MerlinEgalite]

Fixes #480

[QGarchery]

I think I prefer just changing the inequality is _isHealthy into a strict inequality. Comments about 1 WEI of asset feel a bit too much in my opinion

[Rubilmax]

I think I prefer just changing the inequality is _isHealthy into a strict inequality. Comments about 1 WEI of asset feel a bit too much in my opinion

I don't see how it solves the issue: it just shifts the issue by 1. See my reasoning considering a strict inequality:

When borrowing max (borrowed == maxBorrow - 1), your position would be considered healthy with maxBorrow - 1 while the protocol would round up twice and make your position unhealthy (maxBorrow == borrowed)

[MerlinEgalite]

I've created a draft PR #490. Tbh I think it's more difficult to reason about especially in the tests you must tweaks some things to have the behavior you expect which I'm not confortable with...

[QGarchery]

I don't see how it solves the issue: it just shifts the issue by 1

You are right, it doesn't change much. The intent is to signal that the user shouldn't expect to have precision down to the WEI. I still feel like it's not worth adding a big comment just for that: it is putting the focus on a small issue. For example a tiny change in price can change the behavior. If we were to add a comment, just making clear that precision is not down to the WEI should be enough (and it's more general and applies to other computations)

[Rubilmax]

If we were to add a comment, just making clear that precision is not down to the WEI should be enough (and it's more general and applies to other computations)

I'm fine with that too
Where though?

[MerlinEgalite]

You are right, it doesn't change much. The intent is to signal that the user shouldn't expect to have precision down to the WEI. I still feel like it's not worth adding a big comment just for that: it is putting the focus on a small issue. For example a tiny change in price can change the behavior. If we were to add a comment, just making clear that precision is not down to the WEI should be enough (and it's more general and applies to other computations)

Would the following works for you @QGarchery ?
/// @dev The precision of the inequality can vary up to 1 unit of the borrowable token.

[QGarchery]

Would the following works for you @QGarchery ?

Perfect

[MathisGD]

This fix feels a little bit weird to me. We round in favor of the market every time right ? When you borrow and then repay you might need to repay one more asset. When you supply and withdraw you might be able to withdraw one less asset. It's not so surprising to see the "same" thing happening in _isHealthy. Anyway I'm not against the fix but I don't think that it was mandatory at all.