Announcing Gommap

[Gustavo Niemeyer]

Greetings,

The blog post has some interesting details about how it was done and why:

Removing seatbelts with the Go language for mmap support:
http://j.mp/gommap

--
Gustavo Niemeyer
http://niemeyer.net
http://niemeyer.net/blog
http://niemeyer.net/twitter

[nsf]

A very interesting blog post, thanks. Useful package too.

[jimt]

Nice work! This may just solve my issues with the kernel module
wrapper I was writing for V4L (Video For Linux). Once I get my webcam
back, I can test it. I'll let you know how it works out.

[Eleanor McHugh]

On 28 Nov 2010, at 18:40, Gustavo Niemeyer wrote:
> Greetings,
>
> The blog post has some interesting details about how it was done and why:
>
> Removing seatbelts with the Go language for mmap support:
> http://j.mp/gommap

Very nice :)
I can't help but wonder though if there isn't a way of replacing the MMap struct with just a []uint8 derived type... if I can make some time I'll have a hack and see how that works.

Ellie

Eleanor McHugh
Games With Brains
twitter: @feyeleanor
----
raise ArgumentError unless @reality.responds_to? :reason

[Gustavo Niemeyer]

> Very nice :)

Thanks! (to everyone)

> I can't help but wonder though if there isn't a way of replacing the MMap struct with just a []uint8 derived type... if I can make some time I'll have a hack and see how that works.

Yes, I pondered about this alternative too, and technically it would
certainly work that way, but doing so would also force the casting
back to []uint8, and would force keeping two different variables
around: mmap for the methods, and []uint8(mmap) for usage. mmap +
mmap.Data gives exactly that, in a comfortable way, so I opted to go
down that route.

[Russ Cox]

> Yes, I pondered about this alternative too, and technically it would
> certainly work that way, but doing so would also force the casting
> back to []uint8, and would force keeping two different variables
> around: mmap for the methods, and []uint8(mmap) for usage.

Why would you need to say []uint8(mmap)?

Russ

[Gustavo Niemeyer]

> Why would you need to say []uint8(mmap)?

I think I misunderstood the situations in which automatic casting happens.

I assumed that this would fail, for instance, but it doesn't:

type MMap []uint8

func main() {
mmap := MMap([]uint8("testing"))
println(bytes.Index(mmap, []byte("st"))) // <=
}

I recall even testing it, but obviously I must have made a mistake.

I will reimplement gommap with this approach to see how it feels.

[Gustavo Niemeyer]

> I will reimplement gommap with this approach to see how it feels.

Ok, it was quite simple to convert (took less than 10 minutes).

Indeed in a few places additional type casting is needed (e.g.
string(mmap) doesn't work, for instance), and the type can now be
built manually by doing MMap(array), which is both an advantage or a
bug, depending on what kind of trickery someone is doing.
(MMap(array).Protect(PROT_READ) for added fun).

The new version is available through goinstall.. let's see how people
feel about this.