Apollo Client 4: ObservableQuery.reobserve() no longer takes NetworkStatus, permanently rewrites query options
In Apollo Client 3 you could force a one-off network request through a watched query, without the in-flight-request deduplication refetch() does, with query.reobserve({ fetchPolicy: 'network-only', pollInterval: 0, context: { queryDeduplication: false } }, NetworkStatus.refetch), for example inside client.refetchQueries({ onQueryUpdated }). The NetworkStatus.refetch argument made it a disposable request that left the query's own options alone. Apollo Client 4 (checked in 4.3.8) removes the second parameter: reobserve(newOptions?) raises TS2554 'Expected 0-1 arguments'. If you only delete the argument, the call becomes non-disposable: _reobserve runs assign(this.options, mergedOptions), so pollInterval: 0 permanently stops that query's polling (a useQuery with pollInterval quietly stops polling), and fetchPolicy stays 'network-only'. The migration guide doesn't mention this.
In v4 only refetch() (NetworkStatus.refetch) and polling use the disposable path. Public reobserve(opts) always merges the options into the query. Fix: don't pass pollInterval. Make the policy change last for one answer: query.reobserve({ fetchPolicy: 'network-only', nextFetchPolicy: query.options.fetchPolicy, context: { ...query.options.context, queryDeduplication: false } }). After the result, applyNextFetchPolicy puts the original fetch policy back, and updatePolling() keeps the existing interval. One side effect: being non-disposable, it unsubscribes the query's current linkSubscription, so an older request that is still on its way has its answer dropped. That's useful when you want a newer answer to win. The cost is that the context override stays (later polls aren't deduplicated). Verified with a MockedProvider test where the first answer arrives late (it fails with a plain refetch()), and in a browser, where polls still fired 60s later.