`move(table)` does not move the table

[arnetheduck]

Description

import tables
var a, b: Table[int, int]
a = move(b)

a copy is generated even if a move is requested:

    genericSeqAssign((&a__testit_u10.data), b__testit_u11.data, (&NTIkeyvaluepairseq__E8pi9b5QNahsURYzXMjh3qw_));
    a__testit_u10.counter = b__testit_u11.counter;
    unsureAsgnRef((void**)&b__testit_u11.data, NIM_NIL);
    b__testit_u11.counter = 0;

orc moves.

Nim Version

2.0.6/refc

Current Output

No response

Expected Output

No response

Possible Solution

No response

Additional Information

No response

[arnetheduck]

swap(a, b); reset(b) works around :facepalm:

[Araq]

Let me tell you a secret: refc really does not support move semantics. Or destructors. Or sink parameters.

[arnetheduck]

Let me tell you a secret: refc really does not support move semantics. Or destructors. Or sink parameters.

works for seq - don't see why it shouldn't work here?

[juancarlospaco]

refc really does not support move semantics. Or destructors. Or sink parameters.

Maybe that should be made more explicit in the docs ? 🤔

[Araq]

Maybe that should be made more explicit in the docs ? 🤔

Sure but there is a reason we used the version number 2.0 and it defaults to mm:orc. Supporting refc with move semantics is a whack-a-mole game as it was never designed for this.

[ringabout]

@arnetheduck it doesn't work for base cases for seq though

var a, b: seq[int]
a = move(b)

Which generates genericSeqAssign in the c code.

[ringabout]

https://github.com/nim-lang/Nim/pull/22229 made it conversative to move data since 2.0.0. 1.6.x does move tables.

[arnetheduck]

https://github.com/nim-lang/Nim/pull/22229 made it conversative to move data since 2.0.0. 1.6.x does moves tables.

ok, I'm not going mad at least, I was pretty sure I had seen move take effect when investigating poor performance while on 1.6 ;)

move helps us migrate signfificant amounts of code away from shallowCopy and other pure-refc constructs but we cannot do that if it makes the applications significantly slower (which additional copies do) - it would be great if we could address some of these lack-of-move issues for refc in 2.0 (we don't care about other orc:isms such as destructors since these are new features - sink, lent etc are pre-2.0 and pre-orc features that are important migration bridges)