String constructor on a view can mutate parent

[nhz2]

This has happened since #53896

Here is a MWE.

julia> a = Memory{UInt8}(undef, 10);

julia> fill!(a, 0x01);

julia> String(view(a, 1:2))
"\x01\x01"

julia> a
0-element Memory{UInt8}

I expected a to not be changed.

This is causing issues in JuliaWeb/HTTP.jl#1166
And I am also using String(view( in https://github.com/JuliaIO/ZipArchives.jl/blob/5bacade2f71e0c28e2aeba09d21a665ecad131cd/src/reader.jl#L106

[adienes]

I believe this is intended

see also: #32528

[nhz2]

The issue is that view protects the parent from being truncated by String in 1.10, but in 1.11.0-beta1 this is broken.
If this is intended what is the correct way to protect the parent from being truncated?

[vtjnash]

In older versions, the view function also enforced a copy, which seems awkward to expect there. We should possibly check though that the truncation of the Memory object only applies if it is backed by a String (and doesn't make a copy)?

[nhz2]

Maybe the view method that converts a Memory to a Vector could be called something like make_vector to avoid confusion with normal view? The difference would be that the result of view would have a fixed size, and create a copy when passed to String.

[KristofferC]

Maybe #53896 should be reverted / rethought in light of this.

[vtjnash]

We also possibly should not unsafely mutate this Memory in this case. That is somewhat more of an Array feature anyways that you should possibly lose when accessing Memory directly

[nhz2]

If I understand:

julia/src/genericmemory.c

Lines 196 to 207 in c38e7cd

	m->length = 0;
	if (how != 0) {
	jl_value_t *o = jl_genericmemory_data_owner_field(m);
	jl_genericmemory_data_owner_field(m) = NULL;
	if (how == 3 &&
	((mlength + sizeof(void*) + 1 <= GC_MAX_SZCLASS) == (len + sizeof(void*) + 1 <= GC_MAX_SZCLASS))) {
	if (jl_string_data(o)[len] != '\0')
	jl_string_data(o)[len] = '\0';
	if (*(size_t*)o != len)
	*(size_t*)o = len;
	return o;
	}

correctly, there are two possible mutations, setting length to 0, and setting the byte after the end to null. Are String required to always be null terminated?

[nhz2]

Fixed by #54927