Avoid mutation of join slice for separate datasets when joins slice capacity is not yet reached

[fhaifler]

I have encountered a bug, when having two separate SelectDatasets with some common part, they can override one anothers joins. Example datasets (build in order):

• common w/ joins: a, b, c
• derived1 (from common) w/ join: d
• derived2 (from common) w/ join: e

Expected behavior would be to see:

• common joins: a, b, c
• derived1 joins: a, b, c, d
• derived2 joins: a, b, c, e

But the result actually is:

• common joins: a, b, c
• derived1 joins: a, b, c, e
• derived2 joins: a, b, c, e

Therefore derived2 dataset overrides the joins of derived1. This can happen for all additional joins for common parts with number of joins < 2^n. This is consequence of append(ret.joins, jc), which reallocates the data only when cap(ret.joins) is reached.

IMHO reslicing joins to a slice with maxed-out cap should fix this issue (practically enforcing copy-on-write) and (hopefully) shouldn't cause issues elsewhere.

I would be glad if this issue could be fixed soon, so we can continue to use this great library without an issue.

Cheers!

[codecov]

Codecov Report

Merging #261 (606a815) into master (71f3e01) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #261   +/-   ##
=======================================
  Coverage   96.69%   96.69%           
=======================================
  Files          60       60           
  Lines        3302     3302           
=======================================
  Hits         3193     3193           
  Misses         94       94           
  Partials       15       15           

Impacted Files	Coverage Δ
exp/select_clauses.go	94.07% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 71f3e01...606a815. Read the comment docs.

[george-oakling]

this fixes our usecase, would you pls merge it @doug-martin ?