Dart vs FxDart › #38 · Poll a flaky API until first success

Poll a flaky API until first success

FxDart wins async

• range
• toAsync
• map
• peek
• dropWhile
• head

Requirement

An export job's status endpoint is deterministically flaky: the first four polls answer pending, the fifth answers ready. Poll it up to ten times, keep a log of every poll, stop at the first success, and report which attempt won — plus how many polls were actually made. No randomness: the failure count is fixed in the code below, so both versions print the same thing every run.

The FxDart version writes the retry as data: range(1, 11) is the poll schedule, map is the transport, peek records the log, and dropWhile + head is the success policy. Because the chain is lazy and pulled one value at a time, head stops the polling — only five requests are ever made.

polling export-2026-07 (up to 10 attempts):
  poll 1: pending
  poll 2: pending
  poll 3: pending
  poll 4: pending
  poll 5: ready
ready on attempt 5; polls actually made: 5

Side by side

Why they differ

Honestly: the native for loop with a break is short, and nobody would call it wrong. The difference is where the pieces live. In the loop, the attempt budget, the logging, and the success test are all tangled into control flow — change one and you re-read the whole body. In the pipeline each concern is its own named step, so swapping the policy (first success → third success, add a transform, widen the budget) means editing one line. And the laziness guarantee — no polls after the winner — is structural in FxDart, while in the loop it depends on the break being in the right place.