fix: if LEAST_UPPER_BOUND returns null, try GREATEST_LOWER_BOUND

[bcoe]

I'm running into issues adding source-map support to c8, in which LEAST_UPPER_BOUND returns mapping.source === null.

If I loosen up our check so that it falls back to the default behavior of GREATEST_LOWER_BOUND in this edge-case, c8's remapping works like a charm, see:

---

@loganfsmyth can you clarify for me why this change might address my issues? I understand that they underlying cause is most likely that V8's coverage ranges don't directly map to the mappings generated by TypeScript ... but why would LEAST_UPPER_BOUND be returning null?

Reading the documentation, shouldn't it simply return the next closest element that's a greater line number than the element I'm remapping -- I guess I don't fully understand the scenarios that result in LEAST_UPPER_BOUND or GREATEST_LOWER_BOUND returning null.

[bcoe]

I believe, in the case of my testing with c8, it was the GREATEST_LOWER_BOUND in the originalEndPositionFor that was resulting in null data; opting for LEAST_UPPER_BOUND works like a charm ... but I don't fully understand why both wouldn't return an element.

[coreyfarrell]

Sorry I haven't really had a chance to gain a reasonable understanding of source-map modules / API's. What I can say is that I've confirmed that this patch combined with istanbuljs/nyc#1080 resolve some issues which were raised starting at nyc@13.2.0.

[coreyfarrell]

nit: You could just return sourceMap.originalPositionFor({ here then mapping could be const.

[bcoe]

I dug into the source-map codebase, and can speak better to why this logic is necessary:

If you have GREATEST_LOWER_BOUND configured, source-map returns -1 rather than the closest element, if there's no element smaller than the node we're searching for in the tree:

    // we are in termination case (3) or (2) and return the appropriate thing.
    if (aBias == exports.LEAST_UPPER_BOUND) {
      return mid;
    } else {
      return aLow < 0 ? -1 : aLow;
    }

inversely, if LEAST_UPPER_BOUND is configured, source-map returns -1 if there is no element greater than the element being searched for:

    if (aBias == exports.LEAST_UPPER_BOUND) {
      return aHigh < aHaystack.length ? aHigh : -1;
    } else {
      return mid;
    }

---

By trying a search for LEAST_UPPER_BOUND and GREATEST_LOWER_BOUND what we're essentially doing is ensuring that we get the mid point returned to us. This behavior seems more reasonable than completely falling over when generated ranges and source map ranges don't line up well -- as is the case with c8.

There's a similar conversation happening here about trying both approaches:

mozilla/source-map#376