Add UUID v4 generation to the standard library. #6618

IridescentRose · 2020-10-09T01:36:12Z

This is a small addition to the standard library which allows you to create a UUID v4 UUID with an allocator and random number generator.

Example use:

const std = @import("std);

pub fn main() !void {
    const uuid = try std.uuid.newv4(your_allocator, your_random);
    std.debug.print("{}", .{uuid.id});
}

Output:
b29f4b7f-945b-4e55-b95b-b6ebeab198e2

lib/std/uuid.zig

lib/std/std.zig

lib/std/uuid.zig

jedisct1 · 2020-10-09T08:33:22Z

UUIDs must be unique, and accepting an arbitrary RNG doesn't guarantee that. An application may inadvertently use a deterministic or badly seeded RNG.

Do we really want the application to provide the RNG? Not having to do so would also simplify the interface.

lib/std/uuid.zig

IridescentRose · 2020-10-09T08:39:29Z

@jedisct1 How would you propose making it truly unique?

jedisct1 · 2020-10-09T08:42:39Z

@jedisct1 How would you propose making it truly unique?

With crypto.randomBytes(). That wouldn't be "truly unique", but the probability of a UUID collision would be negligible.

data-man · 2020-10-09T08:46:11Z

For comparison: uuid-rs/uuid

v1 - adds the Uuid::new_v1 function and the ability to create a V1 using an implementation of uuid::v1::ClockSequence (usually uuid::v1::Context) and a timestamp from time::timespec.
v3 - adds the Uuid::new_v3 function and the ability to create a V3 UUID based on the MD5 hash of some data.
v4 - adds the Uuid::new_v4 function and the ability to randomly generate a Uuid.
v5 - adds the Uuid::new_v5 function and the ability to create a V5 UUID based on the SHA1 hash of some data.

lib/std/uuid.zig

jsyrjala · 2020-10-09T10:21:46Z

@jedisct1 How would you propose making it truly unique?

With crypto.randomBytes(). That wouldn't be "truly unique", but the probability of a UUID collision would be negligible.

I think there should be methods for both with and without rng parameter. Sometimes gathering entropy in kernel can be slow. E.g app running in a freshly started Docker container. I have seen cases where it takes minutes to gather enough entropy to generate a couple of random UUIDs.

jedisct1 · 2020-10-09T10:36:16Z

Was is with the randomBytes() function?

On Linux, it is using the getrandom() system call which should never block; there's enough entropy gathered when the system starts to provide all the randomness you will ever need later.

On BSDs, it is using arc4random() that never blocks. Same on WASI. If these functions are not available, it is using /dev/urandom (not /dev/random) that doesn't block either.

If something ever blocks, it means that the system is unable to generate random numbers, so it would be unable to generate UUIDs based on random numbers anyway.

jsyrjala · 2020-10-09T11:14:11Z

Was is with the randomBytes() function?

On Linux, it is using the getrandom() system call which should never block; there's enough entropy gathered when the system starts to provide all the randomness you will ever need later.

On BSDs, it is using arc4random() that never blocks. Same on WASI. If these functions are not available, it is using /dev/urandom (not /dev/random) that doesn't block either.

If something ever blocks, it means that the system is unable to generate random numbers, so it would be unable to generate UUIDs based on random numbers anyway.

It was not related to zig. I don't remember exact details, it used /dev/random or some other thing that definitely blocked, though.

daurnimator · 2020-10-09T11:55:38Z

In many situations you don't want or require a syscall for randomness, e.g. on thread Start you would usually seed a threadlocal csprng from the kernel, and then use that in the thread. Similarly for freestanding targets you will need to provide a rand object.

…

On Fri, 9 Oct 2020, 21:36 Frank Denis, ***@***.***> wrote: Was is with the randomBytes() function? On Linux, it is using the getrandom() system call which should never block; there's enough entropy gathered when the system starts to provide all the randomness you will ever need later. On BSDs, it is using arc4random() that never blocks. Same on WASI. If these functions are not available, it is using /dev/urandom (not /dev/random) that doesn't block either. If something ever blocks, it means that the system is unable to generate random numbers, so it would be unable to generate UUIDs based on random numbers anyway. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6618 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAA7HY3EQ3BHE6W7OYASAMTSJ3RTBANCNFSM4SJO74AQ> .

jedisct1 · 2020-10-09T15:42:48Z

#6622 explains my concerns about requiring a Rand parameter in functions needing secure random numbers.

We currently don't have any safe option to use in that context. And if we had one, would we need more?

lib/std/uuid.zig

IridescentRose · 2020-10-10T03:28:13Z

I'll get on this tomorrow

daurnimator

Looks good enough to me now!

Left a couple of nits that don't need to be fixed

Also, should have space after // before comment.

daurnimator · 2020-10-11T01:19:05Z

lib/std/uuid.zig

+    const flip: u128 = 0b1111 << 48;
+
+    //We then set the 4 bits at 48 to 0x4 (0b0100)
+    return Self{ .id = ((gen.int(u128) & ~flip) | (0x4 << 48)) };


there's an extra set of parenthesis here (the outermost pair)

daurnimator · 2020-10-11T01:19:23Z

lib/std/uuid.zig

+            const selector = @truncate(u4, self.id >> shift);
+            shift += 4;
+
+            buf[i] = chars[@intCast(usize, selector)];


no need for an @intCast here: a cast from u4 to usize will be implicit.

daurnimator · 2020-10-11T01:20:56Z

lib/std/uuid.zig

+pub fn format(value: Self, comptime fmt: []const u8, options: std.fmt.FormatOptions, writer: anytype) !void {
+    var buf = [_]u8{0} ** 36;
+
+    var chars = "0123456789ABCDEF";


Should set to undefined here: var chars: *const [16]u8 = undefined;

Or could reorganise to have:

const chars = if ( ..... ) ... else if (....) ... else ...;

andrewrk · 2020-10-16T23:42:00Z

UUIDs must be unique, and accepting an arbitrary RNG doesn't guarantee that. An application may inadvertently use a deterministic or badly seeded RNG.

Do we really want the application to provide the RNG? Not having to do so would also simplify the interface.

Here's a proposal which is intended to address this use case: #6704

Idea here would be the API could be a simple, global, infallible call to gimmeDatUUID() and it would use this thread-local CSPRNG as the backing RNG.

andrewrk · 2020-11-19T01:11:56Z

To be clear, this PR is blocking on #6704

andrewrk · 2020-12-18T03:45:38Z

This is about to get unblocked by #7482

andrewrk · 2020-12-19T05:03:20Z

std.crypto.random is now available to use.

andrewrk · 2020-12-24T09:59:13Z

Upon further reflection, I think this code is appropriate for a third party library rather than the std lib.

daurnimator suggested changes Oct 9, 2020

View reviewed changes

daurnimator reviewed Oct 9, 2020

View reviewed changes

lib/std/std.zig Outdated Show resolved Hide resolved

daurnimator added the standard library This issue involves writing Zig code for the standard library. label Oct 9, 2020

LemonBoy reviewed Oct 9, 2020

View reviewed changes

lib/std/uuid.zig Outdated Show resolved Hide resolved

daurnimator reviewed Oct 9, 2020

View reviewed changes

lib/std/uuid.zig Outdated Show resolved Hide resolved

lib/std/uuid.zig Outdated Show resolved Hide resolved

lib/std/uuid.zig Outdated Show resolved Hide resolved

LemonBoy reviewed Oct 9, 2020

View reviewed changes

lib/std/uuid.zig Outdated Show resolved Hide resolved

LemonBoy reviewed Oct 9, 2020

View reviewed changes

lib/std/uuid.zig Outdated Show resolved Hide resolved

LemonBoy reviewed Oct 9, 2020

View reviewed changes

lib/std/uuid.zig Outdated Show resolved Hide resolved

daurnimator reviewed Oct 10, 2020

View reviewed changes

lib/std/uuid.zig Outdated Show resolved Hide resolved

lib/std/uuid.zig Outdated Show resolved Hide resolved

lib/std/uuid.zig Outdated Show resolved Hide resolved

daurnimator approved these changes Oct 11, 2020

View reviewed changes

Add UUID v4 generation to the standard library.

0d25550

Vexu closed this Dec 23, 2020

Vexu reopened this Dec 23, 2020

uuid: use std.crypto.random, fix format, add test

919f5ab

andrewrk closed this Dec 24, 2020

IridescentRose deleted the uuids branch January 1, 2021 19:22

Luukdegram mentioned this pull request Jan 9, 2024

Implemented std.crypto.random.uuid() via std.rand.Random.uuid() #18494

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add UUID v4 generation to the standard library. #6618

Add UUID v4 generation to the standard library. #6618

IridescentRose commented Oct 9, 2020

jedisct1 commented Oct 9, 2020

IridescentRose commented Oct 9, 2020

jedisct1 commented Oct 9, 2020

data-man commented Oct 9, 2020

jsyrjala commented Oct 9, 2020

jedisct1 commented Oct 9, 2020

jsyrjala commented Oct 9, 2020

daurnimator commented Oct 9, 2020 via email

jedisct1 commented Oct 9, 2020

IridescentRose commented Oct 10, 2020

daurnimator left a comment

daurnimator Oct 11, 2020 •

edited

Loading

daurnimator Oct 11, 2020 •

edited

Loading

daurnimator Oct 11, 2020

andrewrk commented Oct 16, 2020

andrewrk commented Nov 19, 2020

andrewrk commented Dec 18, 2020

andrewrk commented Dec 19, 2020

andrewrk commented Dec 24, 2020

Add UUID v4 generation to the standard library. #6618

Add UUID v4 generation to the standard library. #6618

Conversation

IridescentRose commented Oct 9, 2020

jedisct1 commented Oct 9, 2020

IridescentRose commented Oct 9, 2020

jedisct1 commented Oct 9, 2020

data-man commented Oct 9, 2020

jsyrjala commented Oct 9, 2020

jedisct1 commented Oct 9, 2020

jsyrjala commented Oct 9, 2020

daurnimator commented Oct 9, 2020 via email

jedisct1 commented Oct 9, 2020

IridescentRose commented Oct 10, 2020

daurnimator left a comment

Choose a reason for hiding this comment

daurnimator Oct 11, 2020 • edited Loading

Choose a reason for hiding this comment

daurnimator Oct 11, 2020 • edited Loading

Choose a reason for hiding this comment

daurnimator Oct 11, 2020

Choose a reason for hiding this comment

andrewrk commented Oct 16, 2020

andrewrk commented Nov 19, 2020

andrewrk commented Dec 18, 2020

andrewrk commented Dec 19, 2020

andrewrk commented Dec 24, 2020

daurnimator Oct 11, 2020 •

edited

Loading

daurnimator Oct 11, 2020 •

edited

Loading