fix: regression by reverting fix that returned 206 when first-pos=length in Range header

[laurenceisla]

Related to: #2824 (comment)

[steve-chavez]

@laurenceisla Just so I understand. Once we merge this, the next step would be to do #3007 to fix #2824 ?

[wolfgangwalther]

How about adding a test-case with offset=0 and empty response together with the revert, so that #2824 (comment) can never break again?

[wolfgangwalther]

And probably a test-case for this as well:

Still, the error was happening even when Prefer: count=exact was present but when Range: ... wasn't.

[laurenceisla]

@laurenceisla Just so I understand. Once we merge this, the next step would be to do #3007 to fix #2824 ?

@steve-chavez Not necessarily, but I think it would be ideal and more straightforward to do it that way (note that it's a breaking change).

How about adding a test-case with offset=0 and empty response together with the revert, so that #2824 (comment) can never break again?

And probably a test-case for this as well:

Still, the error was happening even when Prefer: count=exact was present but when Range: ... wasn't.

@wolfgangwalther Good idea, I'll add those.

[laurenceisla]

It looks like ra-data-postgrest uses the Prefer: count=exact header by default.

https://github.com/raphiniert-com/ra-data-postgrest/blob/cc3d412c3789bdd87bfd792f629631d9bdab9f54/src/index.ts#L137-L144

That's why the test I added for only offset=0 did not break for v11.2.1, but I'll leave it just in case.