Thread overview
Handling endianness in Tango's FileConduit
Oct 04, 2008
Nick Sabalausky
Oct 04, 2008
Alexander Pánek
Oct 04, 2008
Nick Sabalausky
Oct 04, 2008
Robert Fraser
Oct 04, 2008
Alexander Pánek
Oct 08, 2008
Nick Sabalausky
October 04, 2008
I've been looking at the Tango docs, but I'm unclear on how IProtocol works, specifically in relation to endianness.

I see stuff about protocols and endianness in the docs, so I assume Tango provides a way to say "The file I'm reading/writing through this conduit uses X type of endieanness, so Tango, make sure that any shorts, ints, etc, that I read/write are adjusted to match the system's native endianness."  If so, how would I adjust the following to tell Tango "This file uses little-endian"?:

auto file = new FileConduit(infilename);
scope(exit) file.close();
auto input = new Reader(file.input);
// Load data through 'input'

BTW, A small side question: Tango's conduits don't have anything like "char[] readNullTerminatedString()", do they? It's trivial enough to implement, but I thought if it was there, I may as well just use it. (I don't think it is there though, right? No big deal, though, just wondering.)


October 04, 2008
Nick Sabalausky wrote:
> I've been looking at the Tango docs, but I'm unclear on how IProtocol works, specifically in relation to endianness.
> 
> I see stuff about protocols and endianness in the docs, so I assume Tango provides a way to say "The file I'm reading/writing through this conduit uses X type of endieanness, so Tango, make sure that any shorts, ints, etc, that I read/write are adjusted to match the system's native endianness."  If so, how would I adjust the following to tell Tango "This file uses little-endian"?:
> 
> auto file = new FileConduit(infilename);
> scope(exit) file.close();
> auto input = new Reader(file.input);
> // Load data through 'input'

tango.io.protocol.EndianProtocol should do that:

auto file = new FileConduit(infilename);
scope(exit) file.close;

auto protocol = new EndianProtocol(file);
auto input = new DerivedReader(protocol);

Note: it seems like this is always assuming the other endianess as your current system runs on.

> BTW, A small side question: Tango's conduits don't have anything like "char[] readNullTerminatedString()", do they? It's trivial enough to implement, but I thought if it was there, I may as well just use it. (I don't think it is there though, right? No big deal, though, just wondering.) 

I don’t think so.
October 04, 2008
"Alexander Pánek" wrote
> Nick Sabalausky wrote:
>> I've been looking at the Tango docs, but I'm unclear on how IProtocol works, specifically in relation to endianness.
>>
>> I see stuff about protocols and endianness in the docs, so I assume Tango provides a way to say "The file I'm reading/writing through this conduit uses X type of endieanness, so Tango, make sure that any shorts, ints, etc, that I read/write are adjusted to match the system's native endianness."  If so, how would I adjust the following to tell Tango "This file uses little-endian"?:
>>
>> auto file = new FileConduit(infilename);
>> scope(exit) file.close();
>> auto input = new Reader(file.input);
>> // Load data through 'input'
>
> tango.io.protocol.EndianProtocol should do that:
>
> auto file = new FileConduit(infilename);
> scope(exit) file.close;
>
> auto protocol = new EndianProtocol(file);
> auto input = new DerivedReader(protocol);
>
> Note: it seems like this is always assuming the other endianess as your current system runs on.

Try PickleProtocol.  This always assumes the data being read/written is big endian.  I don't think there's an equivalent one if the file is in little endian form, but you could always just copy what PickleProtocol does (make an alias depending on native endianness).  In fact, PickleProtocol should probably be changed to allow you to specify the endianness you want.

-Steve


October 04, 2008
"Steven Schveighoffer" <schveiguy@yahoo.com> wrote in message news:gc8ape$2i2r$1@digitalmars.com...
> "Alexander Pánek" wrote
>> Nick Sabalausky wrote:
>>> I've been looking at the Tango docs, but I'm unclear on how IProtocol works, specifically in relation to endianness.
>>>
>>> I see stuff about protocols and endianness in the docs, so I assume Tango provides a way to say "The file I'm reading/writing through this conduit uses X type of endieanness, so Tango, make sure that any shorts, ints, etc, that I read/write are adjusted to match the system's native endianness."  If so, how would I adjust the following to tell Tango "This file uses little-endian"?:
>>>
>>> auto file = new FileConduit(infilename);
>>> scope(exit) file.close();
>>> auto input = new Reader(file.input);
>>> // Load data through 'input'
>>
>> tango.io.protocol.EndianProtocol should do that:
>>
>> auto file = new FileConduit(infilename);
>> scope(exit) file.close;
>>
>> auto protocol = new EndianProtocol(file);
>> auto input = new DerivedReader(protocol);
>>
>> Note: it seems like this is always assuming the other endianess as your current system runs on.
>
> Try PickleProtocol.  This always assumes the data being read/written is big endian.  I don't think there's an equivalent one if the file is in little endian form, but you could always just copy what PickleProtocol does (make an alias depending on native endianness).  In fact, PickleProtocol should probably be changed to allow you to specify the endianness you want.
>
> -Steve

So, if I understand everything right:

- What EndianProtocol does: endianness is always switched.
- What PickleProtocol does: file's endianness is always assumed to be big
endian and converted appropriately.

(I don't see any description of those in the docs, it should probably mention what they do and how they're used.)

If that's all what they do, then seems to me it should work more like this:

1. Rename "EndianProtocol" to "EndianSwitchProtocol" to make it's behavior
more clear. Possibly make it internal to Tango.
2. Create a new "EndianProtocol" that takes an "Endianness" enum of
"LittleEndian" or "BigEndian" (and maybe also "SwitchEndian" and
"NativeEndian") as a ctor parameter, checks the native endianness, and
appropriately adds or doesn't add the "EndianSwitchProtocol" (or just does
the byte-order swapping itself instead of using "EndianSwitchProtocol").
3. Ditch the PickleProtocol (unless there's something else it also does that
I'm not aware of.)

In any case, it looks like step 2 is something I'd currently have to implement myself. Does Tango provide a way to detect the system's endianness?


October 04, 2008
Nick Sabalausky wrote:
> 3. Ditch the PickleProtocol (unless there's something else it also does that I'm not aware of.)

http://www.python.org/doc/2.5.2/lib/pickle-protocol.html
October 04, 2008
Nick Sabalausky wrote:
> 1. Rename "EndianProtocol" to "EndianSwitchProtocol" to make it's behavior more clear. Possibly make it internal to Tango.
> 2. Create a new "EndianProtocol" that takes an "Endianness" enum of "LittleEndian" or "BigEndian" (and maybe also "SwitchEndian" and "NativeEndian") as a ctor parameter, checks the native endianness, and appropriately adds or doesn't add the "EndianSwitchProtocol" (or just does the byte-order swapping itself instead of using "EndianSwitchProtocol").

Please create a ticket for that!

> In any case, it looks like step 2 is something I'd currently have to implement myself. Does Tango provide a way to detect the system's endianness?

Tango doesn’t, but D does with version(BigEndian) or version(LittleEndian).

Also, if you implement it yourself, you can as well contribute it to Tango, so other people don’t have the same problem. :) You can also add a patch/code to the ticket. Would be awesome.

Kind regards, Alex
October 08, 2008
> Nick Sabalausky wrote:
>> 1. Rename "EndianProtocol" to "EndianSwitchProtocol" to make it's
>> behavior more clear. Possibly make it internal to Tango.
>> 2. Create a new "EndianProtocol" that takes an "Endianness" enum of
>> "LittleEndian" or "BigEndian" (and maybe also "SwitchEndian" and
>> "NativeEndian") as a ctor parameter, checks the native endianness, and
>> appropriately adds or doesn't add the "EndianSwitchProtocol" (or just
>> does the byte-order swapping itself instead of using
>> "EndianSwitchProtocol").
>
> Please create a ticket for that!
>
>> In any case, it looks like step 2 is something I'd currently have to implement myself. Does Tango provide a way to detect the system's endianness?
>
> Tango doesn't, but D does with version(BigEndian) or version(LittleEndian).
>

*Smacks forehead* I knew that, and managed to forget :) That was one of the early things that first impressd me about D.

> Also, if you implement it yourself, you can as well contribute it to Tango, so other people don't have the same problem. :) You can also add a patch/code to the ticket. Would be awesome.
>

Good idea. Done:

http://www.dsource.org/projects/tango/ticket/1321

> Kind regards, Alex