Mitigate undefined errors in PValueColumn

[lukesonnet]

Features and Changes

When building the ResultsTable we may pass some default stats if a row is missing for a variation for some reason. We do:

  {variations.map((v, i) => {
    const stats = row.variations[i] || {
      value: 0,
      cr: 0,
      users: 0,
    };
    .BUILD SOME THINGS.
    }

Because we don't populate expected or pValue, if this code were to execute for some edge case, then the PValueColumn code which expects defined values for those fields will throw an error.

This PR fixes that by letting them be undefined and setting them to some default values if needed. The values themselves shouldn't matter since users should be 0 and therefore there should not be enough data to generate anything other than an empty <td>.

I think this is causing a customer reported error and I think that this patch will at least fix the immediate issue.

We may prefer to figure out why this edge case is being created at all and more directly solve the issue, but this may be sufficient to prevent the front-end from throwing errors that lock the user out of the experiment page altogether.

[github-actions]

Your preview environment pr-1025-bttf has been deployed.

Preview environment endpoints are available at:

• https://growthbook-3100-pr-1025-bttf.growthbook.okteto.dev
• https://growthbook-3000-pr-1025-bttf.growthbook.okteto.dev

[bttf]

This addresses the problem, however I'm wondering if it might be better to exit early / render nothing if the above values are missing? Ultimately will defer to your opinion on this but curious to hear what you think.

[lukesonnet]

Yeah I was wondering the same thing, but decided to mostly ape the logic in ChanceToWinColumn. I can test both with my broken MongoDB and see what seems better.