open memoryview object as-is they were bytes

[arthurdarcet]

it supports the same protocol and avoids an unecessary copy

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[arthurdarcet]

I have read the CLA Document and I hereby sign the CLA

[JorjMcKie]

I don't think that this change makes sense.
The base library does ultimately require a memory area to work with, it cannot use memoryview objects directly.
So we would have to convert a memoryview object to bytes before calling MuPDF. This conversion (bytes(mv), respectively mv.tobytes()) however does make a copy and thus negates any assumed advantage.
I think that for a memoryview object mv passing bytes(mv) to the constructor is a negligible effort.

[arthurdarcet]

Could you point me where that copy happens in my fork? The change I proposed works as is, without any extra copy as far as I can see. I'm also not sure how the mv would behave any differently from a bytes object, apart from what type(obj) returns of, hence my proposed change, as that's more or less how the python doc define mem views Le sam. 8 févr. 2025, 14:36, Jorj X. McKie ***@***.***> a écrit :

…

I don't think that this change makes sense. The base library does ultimately require a memory area to work with, it cannot use memoryview objects directly. So we would have to convert a memoryview object to bytes before calling MuPDF. This conversion (bytes(mv), respectively mv.tobytes()) however *does* make a copy and thus negates any assumed advantage. I think that for a memoryview object mv passing bytes(mv) to the constructor is a negligible effort. — Reply to this email directly, view it on GitHub <#4272 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAHNVDRXOZYUL5ROJX3QUYT2OYB7TAVCNFSM6AAAAABWTHBF6WVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNBVGY2DMMJRGI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[JorjMcKie]

@arthurdarcet - sorry, you are right:
I wasn't aware of that the C-code had already been changed to directly support Python's buffer protocol. Therefore all binary formats - bytes, memoryview and bytearray - are readily supported without the need for extra conversions.
Could you please use the isinstance() function to perform the checks - instead of using type(stream)?

[arthurdarcet]

@JorjMcKie of course - just pushed that. I also moved some tests a bit, but logic should still be exactly the same.
I left the assert on line 2933 - but i don't think it's needed? It doesn't hurt, but it will always pass. Let me know if you want me to remove it