Foreign Language characters in filenames [Archive] - Page 2

mariush

19th July 2010, 01:13

And which code page contains the Euro character (€) ? Don't answer, there's several code pages that actually exist and contain it but the point is it's not in ASCII and not all people getting a file with such character may actually have installed that code page.

I won't even get into the new currency symbol for the new Indian Rupee (http://en.wikipedia.org/wiki/File:Indian_Rupee_symbol.svg) for example which is planned to be included in the UTF-8 - the rupee is represented by the Unicode character 20A8 (₨) right now, but it's going to receive another position for the new symbol... the point is code pages is antique technology..

The sensible way is to move all the way to utf-8 which is perfectly fine and can represent if not all, then almost all characters out there. There are free libraries like the IBM's unicode one which would make the job easier but I'm not that good of a programmer to get involved.

ps.. some people work as freelancers therefore receive files with various names from various parts of the world - sometimes included in winrar archives, don't make the assumption that a video is created on one computer and ends its life on the same computer, on the same code page window has running.

Groucho2004

19th July 2010, 03:37

And which code page contains the Euro character (€) ?

Hm, I tried but can't figure out why one would use the Euro symbol in an AVISynth script.

ps.. some people work as freelancers therefore receive files with various names from various parts of the world - sometimes included in winrar archives, don't make the assumption that a video is created on one computer and ends its life on the same computer, on the same code page window has running.

I'm not saying it's perfect and I certainly agree that Unicode AVISynth would be the best way to go but we have to work with what's given. I find it hard to believe that anyone would go through the trouble porting AVISynth (and all plugins in existance) to Unicode.

Gavino

19th July 2010, 10:59

Hm, I tried but can't figure out why one would use the Euro symbol in an AVISynth script.
In a subtitle - just tried it and it works fine for me.

krieger2005

19th July 2010, 20:07

actually i am working as a software developer and know exactly what i'm talking about since i used the libraries behind converting character from codepages to unicode and the way back and so must know the theory behind usage of such libraries. Using Codepages is a possible solution and used decades away. but when you start using Character which are in two different codepages you can't use codepages or must search very long for a codepage which support such character-combinations (for example german: ä and russian я).

Why should someone use such a mix? Just because it's possible!!!! Don't ask why peaple do something - because peaple just do it.

foxyshadis

19th July 2010, 21:57

To avoid recompiling plugins, and since AVISYNTH_INTERFACE_VERSION isn't passed into any constructors, you could either attempt to find the value within the plugin, or have two copies of AddFunction, a char * and a wchar_t * version, the plugin gets whichever environment it tries to use. Kind of nasty, but that's baked code for you. Actually, you could even modify the header to point AddFunction (ThrowError, etc) to an internal AddFunctionW when recompiled, since the filter only gets the functionality if recompiled, but never loses any; that way you don't need to hack in a char * check.

No matter what happens, there will have to be char->wchar conversion functions, because not all plugins are ever going to be recompiled.

IanB

20th July 2010, 00:07

Okay here is a proposal to chew on. Please think this through and post comment if you find issues in the theory.
Define existing internal ansi char* text to now be UTF-8 encoded!
Change Import() to detect BOM prefix for UTF-8 & Unicode input else by default convert from ANSI input using :- MultiBytetoWideChar((AreFileApisANSI() ? CP_ACP : CP_OEMCP, ...)
WideChartoMultiByte(CP_UTF8, ...)Maybe support a magic #!UTF-8 style first line detection to override default behaviour.
Provide a new IScriptEnvironment API Helper :-const wchar* UTF8toWide(const char*);which wraps a MultiBytetoWideChar(CP_UTF8, ...) with an env->SaveString()
Change all the internal filesystemA(name) calls to filesystemW(env->UTF8toWide(name)).
Add some script helper functions for legacy 2.5 plugins that internally use filesystemA(name) calls with input AVS argument strings:-
ShortFileName(string Name) short 8.3 mapped names are defined to be 7bit ascii, but not all file systems support the short name space.
UTF8toANSI(string Name) using AreFileApisANSI() ? CP_ACP : CP_OEMCP to steer the conversion.
UTF8toACP(string Name) using CP_ACP for the conversion.
UTF8toOEMCP(string Name) using CP_OEMCP for the conversion.
UTF8toCodePage(string Name, int CodePage) using tthe user specified codepage number for the conversion.
Notes :-

This is transparent for 7 bit ascii. Yes!

This breaks existing 8 bit codepage behaviour that currently works satisfactorily for 2.5 plugins. Bummer!

This does not hinder a future pure 16bit unicode implementation of Avisynth. Yes!

kemuri-_9

20th July 2010, 00:50

is adding wchar_t * support to AVSValue included in this?

more specifically:
* creating AVSValues from wchar_t *
* AsString() being able to return wchar_t *

IanB

20th July 2010, 03:26

@kemuri-_9,

I did look into extending AVSValue's range of types, but I couldn't see a way of supporting the 2.5 API baked code.

I could add code that internally does WideChartoMultiByte(CP_UTF8, ...) in a new AVSValue(const wchar* s) contructor. And do a MultiBytetoWideChar(CP_UTF8, ...) in a new AsWString() method but that would be pretty heavy on the SaveString storage.

I could be more inclined to add a const char* WidetoUTF8(const wchar*); helper for symmetry.

kemuri-_9

20th July 2010, 03:55

@kemuri-_9,
I could be more inclined to add a const char* WidetoUTF8(const wchar*); helper for symmetry.

this seems fair except this entire concept seems to be putting fairly large strain on env->SaveString().
In cases where only the temporary use of the other version that what was given is required this would just waste memory.

It seems reasonable to add something along the lines of an "AVSString" class that would make the conversion between char* and wchar_t* transparent but stay outside the use of env->SaveString() be added to avisynth's codebase available for plugin devs' use as well.

This would allow plugin devs the use of conversions without having to know how avisynth does it, and be for temporary usage (to avoid unnecessary abuse of env->SaveString())
we have something like this at work and it makes things fairly convenient.

mariush

20th July 2010, 04:00

Just keep in mind that a character in UTF-8 may have up to 4 bytes. Though unusual, I don't think it would be impossible to encounter characters using 3 bytes, like mathematical characters... see this list for all the possible ranges: http://www.unicode.org/Public/UNIDATA/Blocks.txt

I'm saying this because I'm not sure what you mean by "This does not hinder a future pure 16bit unicode implementation of Avisynth. Yes! " and why would you think going the pure 16bit unicode way would be smart - utf-8 seems more sane to me based on what I've read. UTF16 would just force you to use 2 bytes for every char, at least, yet you'd still need up to 4 bytes for one, just like with utf-8.

If you need this for optimizations or something, then utf-32 and 32bit words would seem a better choice... but maybe it's not worth it.

Groucho2004

20th July 2010, 08:23

IanB

20th July 2010, 08:54

This seems fair except this entire concept seems to be putting fairly large strain on env->SaveString().
In cases where only the temporary use of the other version that what was given is required this would just waste memory.

It seems reasonable to add something along the lines of an "AVSString" class that would make the conversion between char* and wchar_t* transparent but stay outside the use of env->SaveString() be added to avisynth's codebase available for plugin devs' use as well.

This would allow plugin devs the use of conversions without having to know how avisynth does it, and be for temporary usage (to avoid unnecessary abuse of env->SaveString()) we have something like this at work and it makes things fairly convenient.
I'm not sure what you mean by an "AVSString" class, but it seems like you are suggesting something along the lines of the STL String class or the related IOstream class. If so why reinvent the wheel. STLPort has a real beast of an extended implementation if you need such.

The current SaveString model is a little wasteful but it is a fair trade for simplicity and stability. An alternative would be along the lines of strdup(), where the programmer has to eventually remember to do a free() [and they always forget :devil: ] Given char* strings are already successfully using the SaveString model doing wchar* strings a similar way is not a great stretch.

On my quick pass through the code to scope out the work needed I did not find a need for a general WidetoUTF8() feature, hence it was forgotten.

Just keep in mind that a character in UTF-8 may have up to 4 bytes. Though unusual, I don't think it would be impossible to encounter characters using 3 bytes, like mathematical characters... see this list for all the possible ranges: http://www.unicode.org/Public/UNIDATA/Blocks.txt

I'm saying this because I'm not sure what you mean by "This does not hinder a future pure 16bit unicode implementation of Avisynth. Yes! " and why would you think going the pure 16bit unicode way would be smart - utf-8 seems more sane to me based on what I've read. UTF16 would just force you to use 2 bytes for every char, at least, yet you'd still need up to 4 bytes for one, just like with utf-8.

If you need this for optimizations or something, then utf-32 and 32bit words would seem a better choice... but maybe it's not worth it.
Actually the UTF-8 encoding scheme can have up to 6 bytes for a given code point, this is not really a problem, as all code points >127 are represented by bytes all with the 128 bit set. The only visible issue is with the line position pointer in the parser used for reporting script errors.

Fragment of code from a UTF-8 encoder if (ch < 128) {
bt1 = ch;
}
else if (ch <= 2047) {
bt1 = (BYTE)(192 + (ch / 64));
bt2 = (BYTE)(128 + (ch % 64));
}
else if (ch <= 65535) {
bt1 = (BYTE)(224 + (ch / 4096));
bt2 = (BYTE)(128 + ((ch / 64) % 64));
bt3 = (BYTE)(128 + (ch % 64));
}
else if (ch <= 2097151) {
bt1 = (BYTE)(240 + (ch / 262144));
bt2 = (BYTE)(128 + ((ch / 4096) % 64));
bt3 = (BYTE)(128 + ((ch / 64) % 64));
bt4 = (BYTE)(128 + (ch % 64));
}
else if (ch <=67108863) {
bt1 = (BYTE)(248 + (ch / 16777216));
bt2 = (BYTE)(128 + ((ch / 262144) % 64));
bt3 = (BYTE)(128 + ((ch / 4096) % 64));
bt4 = (BYTE)(128 + ((ch / 64) % 64));
bt5 = (BYTE)(128 + (ch % 64));
}
else if (ch <=2147483647) {
bt1 = (BYTE)(252 + (ch / 1073741824));
bt2 = (BYTE)(128 + ((ch / 16777216) % 64));
bt3 = (BYTE)(128 + ((ch / 262144) % 64));
bt4 = (BYTE)(128 + ((ch / 4096) % 64));
bt5 = (BYTE)(128 + ((ch / 64) % 64));
bt6 = (BYTE)(128 + (ch % 64));
}

IanB

20th July 2010, 09:51

One important thing to consider is that there are text editors out there which don't necessarily write a BOM when saving to UTF-8 or even UTF-16, either by default or user-configured.

Most advanced editors (UltraEdit, EMEditor, Unipad) scan a document upon opening to determine the encoding if they don't find a BOM.

So, the absence of a BOM doesn't necessarily mean that the file has 8-bit (Ansi) encoding.Both 16bit Unicode variants {LE, BE} are always easy to recognise by the null bytes in each pair of bytes even without a BOM.

Distinguishing UTF-8 from an 8 bit ansi codepage encoding is very difficult without the BOM. On first glance both are streams of bytes [1..255]. So that doesn't help. However UTF-8 will match one of these patterns :-0xxxxxxx
110xxxxx 10xxxxxx
1110xxxx 10xxxxxx 10xxxxxx
11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxxSo a file not matching this cannot be UTF-8, but the opposite is not true. A file matching this can still be an 8 bit ansi codepage encoding. And given we want to assume a BOMless file is an 8 bit ansi codepage encoding we are a little stuck. Strictly speaking even the BOM could be a valid 8 bit ansi codepage encoding but such a file cannot be a valid Avisynth script.

Thus my idea for a magic first line :-Maybe support a magic #!UTF-8 style first line detection to override default behaviour.

Groucho2004

20th July 2010, 10:33

Maybe this helps with the encoding detection:
http://www.codeproject.com/KB/recipes/DetectEncoding.aspx

I did some experiments with MLang a long time ago and the results were pretty good.

IanB

20th July 2010, 11:04

All Fire Trucks are red!

Your car is red.

Therefore your car is a Fire Truck.
________________________________

All UTF-8 files match said pattern!

Your file matches said pattern.

Therefore your file is UTF-8

:devil:

Gavino

20th July 2010, 11:25

Actually the UTF-8 encoding scheme can have up to 6 bytes for a given code point, this is not really a problem, as all code points >127 are represented by bytes all with the 128 bit set. The only visible issue is with the line position pointer in the parser used for reporting script errors.
What about the string handling functions in the script language, especially those that refer to index or length? For example, FindStr, LeftStr, RightStr, MidStr (or even StrLen). Does it all come out in the wash, transparently to the user?

Groucho2004

20th July 2010, 11:36

mariush

20th July 2010, 13:40

RFC 3629 (http://tools.ietf.org/html/rfc3629) restricts UTF-8 to 4 Bytes.

Theoretical, yes, UTF-8 is up to 6 bytes but just like with the x64 architecture which only uses 44 bits out out 64 bits of memory addressing afaik, it looks like for a few years at least it's safe to assume it will only be up to 4 bytes.

UTF-8 files are not required to have a BOM (which is 0xEF, 0xBB, 0xBF for UTF-8).. see this: http://www.unicode.org/faq/utf_bom.html#bom1

You're supposed to detect it the file is UTF-8 or not and that doesn't work quite well.. as you can see by looking on Youtube at all those clips showing how when you write "bush hid the facts" (http://www.youtube.com/watch?v=FueE3WqHNl4) in Notepad the saved result is garbled when viewed back.

Well, it's just bad unicode detection in windows... these pages explain how the encoding is detected (the blogger works at Microsoft at fonts or something like that):
http://blogs.msdn.com/b/michkap/archive/2006/07/11/662342.aspx
http://blogs.msdn.com/b/michkap/archive/2008/03/25/8334796.aspx
http://blogs.msdn.com/b/michkap/archive/2007/04/22/2239345.aspx

I'm not sure if something like #!UTF8 would be wise, I don't often see something NEGATING...

My first thought would be at something like EnableUTF8 = TRUE at the beginning of the script (which would be by default FALSE to keep compatibility with older plugins and be TRUE by default on a later version, 3.0 for example)

Or maybe just having to use the .AVSU extension instead of AVS (just like Windows uses A and W at the end of functions to make distinction)... There's no problem in Windows having more than 3 characters in extension

Or maybe autodetect if the filename is in the form of.. let's say Script.UTF8.avs or Script.UTF-8.avs

ps.

So a file not matching this cannot be UTF-8, but the opposite is not true. A file matching this can still be an 8 bit ansi codepage encoding. And given we want to assume a BOMless file is an 8 bit ansi codepage encoding we are a little stuck. Strictly speaking even the BOM could be a valid 8 bit ansi codepage encoding but such a file cannot be a valid Avisynth script.

It doesn't work quite like that. The standard says that if you encounter an error or a notation for a yet undefined character, it's up to you... you're supposed to either silently ignore the bad bytes, on display show ? or the ? reversed (the Spanish "?") and copy the bad bytes as they are if it's not shown... I can't find exactly in what document this was said (I remember it from yesterday) but I can paste a quote from the specs:

(page 94, http://www.unicode.org/versions/Unicode5.2.0/ch03.pdf#G28070)

Although a UTF-8 conversion process is required to never consume well-formed subsequences
as part of its error handling for ill-formed subsequences, such a process is not otherwise
constrained in how it deals with any ill-formed subsequence itself. An ill-formed
subsequence consisting of more than one code unit could be treated as a single error or as
multiple errors. For example, in processing the UTF-8 code unit sequence <F0 80 80 41>,
the only formal requirement mandated by Unicode conformance for a converter is that the
<41> be processed and correctly interpreted as <U+0041>. The converter could return
<U+FFFD, U+0041>, handling <F0 80 80> as a single error, or <U+FFFD, U+FFFD,
U+FFFD, U+0041>, handling each byte of <F0 80 80> as a separate error, or could take
other approaches to signalling <F0 80 80> as an ill-formed code unit subsequence.

ps2 ... yeah I found that in Wikipedia actually : http://en.wikipedia.org/wiki/UTF-8#Invalid_byte_sequences

RFC 3629 states "Implementations of the decoding algorithm MUST protect against decoding invalid sequences."[7] The Unicode Standard requires decoders to "...treat any ill-formed code unit sequence as an error condition. This guarantees that it will neither interpret nor emit an ill-formed code unit sequence."
[..]
Popular replacements are:

* The replacement character '�' (U+FFFD)
* The '?' or '¿' character (U+003F or U+00BF)
* The invalid Unicode code points U+DC80..U+DCFF where the low 8 bits are the byte's value.
* Interpret the bytes according to another encoding (often ISO-8859-1 or CP1252).

Replacing errors is "lossy": more than one UTF-8 string converts to the same Unicode result. Therefore the original UTF-8 should be stored, and translation should only be used when displaying the text to the user.

IanB

20th July 2010, 14:37

What part of "cannot detect UTF-8 in the presence 8 bit ANSI code page encoding" do you not understand. Every possible UTF-8 files is also a possible 8 bit ANSI code page file. All the links and references so far pointed note this dilemma.

Unlike my previous facetious expression, the following logic is correct :-

All Fire Trucks are red!
Your car is NOT red.
Therefore your car is NOT a Fire Truck.

i.e. We can easily prove a file is NOT UTF-8 encoded, but we cannot prove that it is.

#!UTF-8 as in #!/bin/sh and #!/usr/bin/perl from Unix.

Parsable constructs like EnableUTF8 = TRUE are not on. All the processing has to be done as Import() reads the contents of the file into memory and before it is Eval()'d by the parser. I'm happy to drop the exclamation point, '!', and just compare the start of the byte stream with something like "#UTF8\r\n".

String handling functions in the script language are gonna be screwed. Document the existing functions as dealing with 8 bit bytes [1..255]. Add new functions that are multi byte aware.

Magic filenames are generally a bad idea and are not really the business of Avisynth.dll, they are more the business of the host application.

mariush

20th July 2010, 15:00

#!UTF-8 as in #!/bin/sh and #!/usr/bin/perl from Unix.

Avisynth is not a Linux application, it's Windows only, so I don't understand why you'd want to import a Unix/Linux concept to a Windows audience. Windows users are not familiarized to that notation.

I agree it's very hard to make distinction between UTF-8 and ANSI pages, that's why it's better to have some sort of tag that tells you it's UTF8. But you shouldn't have a tag that says it's NOT utf-8 ( C/C++ programmers without Unix/Linux may think the ! is NOT).

It should be something introduced as optional at first to mantain compatibility with older plugins until they're updated and then at a later version be enforced.

All Fire Trucks are red!
Your car is NOT red.
Therefore your car is NOT a Fire Truck.

i.e. We can easily prove a file is NOT UTF-8 encoded, but we cannot prove that it is.

Not all fire trucks are red... see http://www.google.ro/images?q=airport%20fire%20truck

http://www.oakhillfire.org/images/aE301.jpg
http://lh5.ggpht.com/_C9zOzLEte2o/Sg6c-N9GMFI/AAAAAAAACI8/RkedMOS4s90/purple-fire-truck.jpg
http://cityofsatsuma.com/images/FireTruckWhiteNov2008.jpg
http://www.indianvalley.net/veterans%20day%202008/hires/ryan/yellow_fire_big-re.JPG

Gavino

20th July 2010, 17:05

Change Import() to detect BOM prefix for UTF-8 & Unicode input else by default convert from ANSI input using :- MultiBytetoWideChar((AreFileApisANSI() ? CP_ACP : CP_OEMCP, ...)
WideChartoMultiByte(CP_UTF8, ...)Maybe support a magic #!UTF-8 style first line detection to override default behaviour.
Note that Import is not the sole route for getting text into Avisynth. Applications using the library interface can also use Eval - for example, I believe AvsP uses Eval instead of Import. Does your proposal mean that such applications will in future have to pass a UTF-8 string to Eval instead of an ANSI one?
String handling functions in the script language are gonna be screwed. Document the existing functions as dealing with 8 bit bytes [1..255]. Add new functions that are multi byte aware.
So some existing scripts (using these functions on strings with 8-bit ANSI) will stop working?

henryho_hk

22nd July 2010, 11:11

Groucho2004

22nd July 2010, 11:29

If your NTFS did not have "8.3 filename generation" disabled, every file has a "secondary" filename which is non-unicode-local-codepage-safe. Do a "dir /x" under a command prompt and you can find them. This is the interim solution I have been using for years to load such files.

Disabling "8.3 filename generation" is a popular tweak so a programmer cannot assume that the short file name exists. Apart from that, dealing with 8.3 is very messy and error prone.

krieger2005

25th July 2010, 10:32

What do you think about this procedure.

1. Detecting of File-Coding can be done in two ways:
a. Reading the BOM
b.if the BOM does not exist, search for Character which mostly exist in Scripts: ".", "(", ")", "{", "}", "=". Determinate based on this Detection the File-Coding (BYTE-CODING, UTF16BE/LE, UTF32BE/LE).
c. if no file-coding could be detected assume BYTE-CODING.

BYTE-CODING can mean UTF-8, but also Coding in some Codepage. Since the Coding in some Codepage means that the Charset can not be interpreted without knowing the Codepage (so the Loading on an external PC can fail) best solution would be to interpret BYTE-CODING allways as UTF-8. For Scripts which are saven in a Codepage an extra Command could tell the Avisynth-Parser which Codepage is used.

Based on the Filecoding the Content should be converted into UTF-8, so that all internal processing is done on UTF-8 char* streams. This step make sure, that
1. Existing Plugins can be used as before since they still work on char* streams
2. Plugins which should support Widechar Character can be sure which type of Intput Stream they get and convert it into they own needed Format - For example mpeg2source could convert the Path into UTF16 (which is used by Windows XP and newer). Still it would be nice for People who implement the Plugin if Avisynth would support Functions to Convert an UTF8-Stream into "what they need".

This is my propose

Gavino

25th July 2010, 13:09

if the BOM does not exist, search for Character which mostly exist in Scripts: ".", "(", ")", "{", "}", "=". Determinate based on this Detection the File-Coding (BYTE-CODING, UTF16BE/LE, UTF32BE/LE).
How practical/reliable is this? Can you give a precise algorithm?
best solution would be to interpret BYTE-CODING allways as UTF-8.
This is not backwards compatible - some existing scripts (using 8-bit ANSI strings) will stop working.
For Scripts which are saven in a Codepage an extra Command could tell the Avisynth-Parser which Codepage is used.
Depends what you mean by 'command'. At best, it would be some simple directive outside the script language.
As IanB said in post #74, "Passable [I think he meant 'parsable'] constructs like EnableUTF8 = TRUE are not on. All the processing has to be done as Import() reads the contents of the file into memory and before it is Eval()'d by the parser."
... the Content should be converted into UTF-8, so that all internal processing is done on UTF-8 char* streams. This step make sure, that
1. Existing Plugins can be used as before since they still work on char* streams
But existing plugins expect to see 8-bit ANSI characters in strings, not UTF-8. Filenames, for, example, would need to be handled differently.
Conversion to UTF-8 is part of IanB's proposal, and as he recognised "This breaks existing 8 bit codepage behaviour that currently works satisfactorily for 2.5 plugins."

IanB

25th July 2010, 13:42

Note that Import is not the sole route for getting text into Avisynth. Applications using the library interface can also use Eval - for example, I believe AvsP uses Eval instead of Import. Does your proposal mean that such applications will in future have to pass a UTF-8 string to Eval instead of an ANSI one?

So some existing scripts (using these functions on strings with 8-bit ANSI) will stop working?
And yes, you stepped on the landmines as well.

I have been avoiding spoon feeding this discussion too much because I am hoping someone will come up with some ideas that do not paint one into a corner. Just about every way I look at this something always comes unglued somewhere.

My first thought about the API text was existing calls as ANSI, add new calls for UTF8, but that has problems.

And yes I really, really don't want existing scripts or applications to stop working.

Thoughts?

@krieger2005,

UTF16BE/LE, UTF32BE/LE are dead easy, I do not have any difficulties with 16 or 32 bit file encoding.

The hard part is 8 bit byte encodings.

The state of play has to be 8 bit ansi codepage encoding has priority.

That is what works now and should continue to work in the future.

UTF-8 is probably going to have to be externally identified. Either by a BOM, some reserved opening sequence on the first line of the file or some other devious scheme.

krieger2005

25th July 2010, 13:48

How practical/reliable is this? Can you give a precise algorithm?
This algorithm is used when detecting HTML/XML Content Coding, but here the Parser search for "<".

This is not backwards compatible - some existing scripts (using 8-bit ANSI strings) will stop working..
ANSI still have the first 7 Bits in common with ASCII and UTF-8. And all the scripts i know does not use special character above the value of 127. And even for this case: it would be practicable to Add the Codepage-Command, so that the Avisynth Parser can convert it internally to UTF-8.

Depends what you mean by 'command'. At best, it would be some simple directive outside the script language
I thought about an Avisynth Command like SetMemoryMax (for Example: UseCodepage(ANSII). I don't think an external derective is a good solution because when people share scripts and one have set this external derective and the other one not it could result in problems reading Scripts.

But existing plugins expect to see 8-bit ANSI characters in strings, not UTF-8. Filenames, for, example, would need to be handled differently.
Do they expect ANSI? Or do they just take what they get? I mean, plugins, which get Unicodestrings have to be updated in any case. For example mpeg2source must call the Widechar Functions to open the File.

EDIT: Sorry forgot the Algorithm:
1. Detect BOM (i will not describe this)
2. if BOM couldn't be detected:
*char_codes = ".{}()=";
for(int i =0; i < min(filesize,50); i++)
b = read_byte_from_file
if(strchr(char_codes, b) != NULL)
look if previous byte was a 0 // UTF 16LE ??
look if next byte is a 0 // UTF16BE ??
if(UTF16LE)
look if the three previous byte are 0 // UTF32LE
if(UT16BE)
look if the three next bytes are 0

If the script contains on of the searched character then the algorithm can determinate in 100% the correct filecoding. I Assume that 99.9% of all script contain one of the characters above. So this algorithm will work for 99.9% of all Files.

krieger2005

25th July 2010, 15:10

The hard part is 8 bit byte encodings.

Ah, i forgot this step... in my program i had the fallback to iso8559-1. I made the Parser for HTML-Files. Additionally i ran the ICU-Charset Detector and used a high relibiality to decide to trust the Detector or not.

But, as i stated above i prefer the Definition of the used Charsed inside the File istead definening somewhere outside something. This way it get just the same as HTML does in their Files.

There is still one difference. I have heard here from people that they use Codepages so the Script works in foreign Languages. Others use ANSI. But: What part of the Script use Characters above the 127 Bounds (and not in comments)? Based on the Answer of this Question, i think you can decide to Support ANSI as Fallback or just suppose the Script to be an UTF-8 Script.

kemuri-_9

25th July 2010, 16:04

And yes, you stepped on the landmines as well.

I have been avoiding spoon feeding this discussion too much because I am hoping someone will come up with some ideas that do not paint one into a corner. Just about every way I look at this something always comes unglued somewhere.

My first thought about the API text was existing calls as ANSI, add new calls for UTF8, but that has problems.

And yes I really, really don't want existing scripts or applications to stop working.

Thoughts?

the least evil way of having this work as i see it is to have an avisynth variable control what character encoding Eval() is expected to be given
(outside of Import's currently planned handling, as import will have the file encoding to handle the situation)

something similar to the other global variables like OPT_UseWaveExtensible, except that this new variable can't be set within scripts to avoid issues that would bring.

something like

AVSValue y = AVSValue( true );
env->SetVar( "$OPT_EvalIsUTF8", y );

would trigger Eval to expect UTF8 strings instead of ANSI and the value is defaulted to false.

this would technically complicate Import though, as it would need to handle dealing with this variable...

Groucho2004

25th July 2010, 16:19

Maybe I'm missing something but why would we need UTF-8 at all? What about either 8 bit Ansi or UTF-16 as choices?

The advantage would be that both encodings are easily identified, whether there is a BOM in UTF-16 or not.

mariush

25th July 2010, 17:41

Notepad and most editors, when you select Unicode at Save as, they'd default at UTF-8. You would have to do 2-3 extra clicks to select on purpose UTF-16.
Once you say your app support Unicode, people would naturally assume you support UTF-8 too, so they'd be constantly surprised if their scripts fail. So you'd have to support UTF-8 anyways.

kemuri - you could also use an environment variable for that, I think.

ps. Would installing two libraries be possible, for example avisynth.dll and avisynthw.dll for unicode ? maybe have some "Stub" or whatever it's called that would then load the appropriate dll depending on how it's initialized?

ps2. Well ok, maybe not Notepad, which has ASCII, then Unicode, then Unicode Big Endian and then UTF-8, at least on my Windows 2003, but other editors have UTF-8 first, like for example the popular Ultraedit:

http://savedonthe.net/image/648/uedit.png

Groucho2004

25th July 2010, 17:50

Once you say your app support Unicode, people would naturally assume you support UTF-8 too, so they'd be constantly surprised if their scripts fail.

Well, that 's just a matter of documentation, isn't it? You don't see lots of people surprised that their car's engine is broken because they put diesel in it instead of petrol (that's gas for our American readers), right?

mariush

25th July 2010, 18:27

When you have a 15-25 thousand car and chances of losing it for a day, you're more careful about that sort of thing. However, even so it will still happen to screw up.

People won't be as careful with a script as with a car, and for lots of people it would be routine, automatism, and they'd expect for something to work as it works with other programs - most won't even read the documentation, as I'm sure you didn't read the manual for Windows or other programs you have first. People will just make mistakes a lot and will be pissed and will fill the forum and most won't even understand the difference between UTF-8 and UTF-16 and won't care to learn - they just want to convert a video.

I've been reading Raymond Chen's blog (http://blogs.msdn.com/b/oldnewthing/) for years (he's a guy that worked at Microsoft on Windows 95 and still works there) and I highly recommend you browse the posts in the History section... I've learned to appreciate the extend Microsoft went to preserve compatibility with lots of things as operating systems improved and often found some things I thought to be wrong to be actually quite smart and appropriate.

I've also found out that we often make assumptions that are not really true in the real world - open source programmers and small developers don't have access to millions of people using the software to know how they think so just assuming they won't use or that they'll learn not to use UTF-8 is bad and won't work in reality.

Gavino

25th July 2010, 19:04

What part of the Script use Characters above the 127 Bounds (and not in comments)?
String literals, of course.
Especially (but not limited to) filenames - which is what I thought was the main point of this thread (or at least the starting point).

krieger2005

25th July 2010, 19:38

String literals, of course.
Especially (but not limited to) filenames - which is what I thought was the main point of this thread (or at least the starting point).
Exactly - Main Point of this thread are Filenames. But who use a script with a filename multiple times? I for myself write for every Movie another script. So how many people use such scripts out there (PS: I forgot Filenames in Imports... okey)?

This is the reason why i asked if it is really usefull to be backward compatible.

The reason why i suggest UTF8 was (i does not know the internals of Avisynth), that i thought, that avisynth serve the Data of the script to the Plugins after it has parsed it. So the main point is: how to should save avisynth the data internally in the parsing step? Support two formats and supporting two interfaces is one possiblity. But why should one do this?

The Benefit of UTF8 is, that it is represented character by a bytestream. So, every old Plugin can be still used with the old interface without a recompile (no wchar, no int for UTF16 character, simply char*).

But maybe i misunderstood the point how avisynth works internally.

kemuri-_9

25th July 2010, 20:16

kemuri - you could also use an environment variable for that, I think.

no, that's asking to get shot with a handgun:
a user could have happened to set the environment variable to 'true' and then they use a program that isn't up-to-date in this regard expecting to use ASCII strings as things stand now.
But due to the environment variable, avisynth will be expecting UTF8 strings.
this would very likely cause unpredictable results and possibly even crashes.

don't even say 'use the registry' either, that's asking to get blasted with a shotgun.

If this route is actually taken, Import() would be required to do something like
1) backup the current value of the variable, whatever it is.
2) read in the file, detecting the encoding as necessary
3) set the variable to be false for the normal ASCII files, otherwise set to true and use UTF-16 -> UTF-8 conversion before passing to Eval(), if necessary.
4) restore the backed-up variable.

doing this with environment variables is ugly especially since they are string values.
the registry is unusable as it can break when multiple instances of avisynth are being used simultaneously.

this is why i proposed to just keep this handling localized within avisynth.

Gavino

25th July 2010, 20:26

But who use a script with a filename multiple times? I for myself write for every Movie another script.
I get your point, but the non-ASCII characters could be in a folder name that is used for many movies. Or (as you say) in an Import of common script code. Also, I have many scripts that I use over and over again for playing (eg a specific combination of clips) rather than encoding.
The reason why i suggest UTF8 was (i does not know the internals of Avisynth), that i thought, that avisynth serve the Data of the script to the Plugins after it has parsed it.
That's correct (assuming by 'data' you mean the parameters passed to the plugin function).
The Benefit of UTF8 is, that it is represented character by a bytestream. So, every old Plugin can be still used with the old interface without a recompile (no wchar, no int for UTF16 character, simply char*).
But if the plugin doesn't 'know' that the string is UTF-8, it won't necessarily do the right thing when using the data. For example, if it is a filename, a different procedure must be used to open it.

henryho_hk

27th July 2010, 07:03

Alternate proposal (similar to shell script headers) for file format:

1) Support 8-bit local codepage and UTF-8 only
2) BOM is optional and ignored (it is optional for UTF-8 plain text files anyway)
3) All UTF-8 encoded AVS files must have a special comment like "#! UTF-8" at the very first line. If it is not found, the file is assumed to be in 8-bit local code page encoding (irrespective of BOM).

99.999% full backward compatibility with existing scripts and easy to implement.

LoRd_MuldeR

21st May 2011, 14:02

Doesn't look like that thread ever came to final conclusion. And even if it eventually will, Unicode (UTF8) support will only be in new versions, not in the old 2.5.x series, right?

So I guess I will stick with the VfW interface for now, which at least should support Unicode names...

I know that this won't work if Unicode strings are used inside the Avisynth script, but at least I can open the AVS file, if the name of the AVS file itself contains Unicode characters.

(Still I have no clue why VFW on my system apparently opens such files successfully, but then reports that there are no streams)

kemuri-_9

21st May 2011, 15:06

indeed, avisynth does not support utf-8 within scripts and API calls, this includes the basic 'Import(file.avs)' call necessary to load existing scripts into avisynth via the API.

when support does come, it is very likely that it will not be backported to earlier releases either.

you could always work around the vfw interface not failing for when the file doesn't exist by doing a _waccess( filepath, 0 ) check beforehand.

LoRd_MuldeR

21st May 2011, 15:14

indeed, avisynth does not support utf-8 within scripts and API calls, this includes the basic 'Import(file.avs)' call necessary to load existing scripts into avisynth via the API.

But what happens in the case when we have an AVS file that does not contain any Unicode (non-Latin1) characters, but the AVS file itself has a Unicode (non-Latin1) name?

I currently won't be able to import that script file with Avisynth' native API, but I am able to open the file with AVIFileOpenW(). Does Avisynth' VFW wrapper fail internally in that case?

At least I have not been able to successfully load a simple AVS file with Unicode-characters in the file name this way...

you could always work around the vfw interface not failing for when the file doesn't exist by doing a _waccess( filepath, 0 ) check beforehand.

Currently I check for the existence/accessibility manually with wfopen(), but this way it's more convenient. Thanks for the hint!

kemuri-_9

21st May 2011, 16:35

But what happens in the case when we have an AVS file that does not contain any Unicode (non-Latin1) characters, but the AVS file itself has a Unicode name?

I currently won't be able to import that script file with Avisynth' native API, but I am able to open the file with AVIFileOpenW(). Does Avisynth' VFW wrapper fail internally in that case?

At least I have not been able to successfully load a simple AVS file with Unicode-characters in the file name this way...

from what i read in the codebase, the (simplified) VFW wrapper flow goes:

entry -> Load( OLEStr, mode )

unchecked convert OLEStr -> char *filename (native/oem codepage - not utf-8)
copy filename to a member variable
...
IScriptEnvironment->Invoke( "Import", filename )

... same code flows as AVS API.

so the only difference is that the VFW does the conversion from utf-16 filename to a native/oem codepage filename itself.

It's Import() that throws an error if the file doesn't exist, and it primarily uses xxxxxA versions of the windows API (avisynth is compiled with MBCS after all).

So if it is working with utf-16 filenames that do not map to the native/oem codepage, that's quite surprising.

LoRd_MuldeR

21st May 2011, 17:50

Yup, that's clearly the reason why it doesn't work with Unicode (non-Latin1) file names, even when AVIFileOpenW() initially succeeds. I think the only workaround at this time is converting the file names to "short" names and passing the short ones AVIFileOpenW(). Short names usually survive the conversion to the local 8-Bit Codepage. It's not nice and it's not guaranteed to work (AFAIK the MSDN doc nowhere says that "short" names can't contain Unicode chars, but it does say that files are not guaranteed to have a short name), but it will probably save us most of time...

kemuri-_9

21st May 2011, 19:00

LoRd_MuldeR

21st May 2011, 19:59

Using shortnames is not likely to work either:

Import uses GetFullPathName or SearchPath (which is used depends on if there's a path delimiter in the filename) to get the fullpath/filename of the script file.
these will error if the filepath contains characters outside of the native/oem codepage.

GetFullPathName() is one of the things that can prevent the "pass short path name" workaround from working, I know :rolleyes:

However with Avisynth it works for me here, passing the "short" version of a fully-qualified path:
avs2wav v1.2 [May 21 2011]
by Jory Stone <jcsston@toughguy.net>, updates by LoRd_MuldeR <mulder2@gmx.de>

Input: G:\ViDeOz\Music\Володимирович 菅直人 κλασικής.avs
Output: Dump.wav

Checking Avisynth... Done
Analyzing input file... Done
Opening output file... Done

[Audio Info]
TotalSamples: 11391413
TotalSeconds: 258
SamplesPerSec: 44100
BitsPerSample: 16
Channels: 2
AvgBytesPerSec: 176400

Dumping audio data, please wait:
0/11391413 [0%]
11391413/11391413 [100%]

All samples have been dumped. Exiting.