FreeDOS/docs/nls.txt

192 lines
8.1 KiB
Plaintext

Current version: $Id: nls.txt 59 2000-08-06 04:42:32Z jimtabor $
This document describes all aspects of the implementation
of NLS in the FreeDOS kernel -- 2000/06/16 ska
Note:
At this time this document contains only an overall description
of how the FreeDOS NLS works; detailed implementation details are
found in HDR\NLS.H and KERNEL\NLS_LOAD.C. When the FreeDOS developers
finally adopted the current scheme, the larger comments of both
files will be merged into a single document -> this file.
= TOC
= Capabilites of the current implementation.
Tested is:
+ DOS-38 - Get Country Information
+ DOS-65-[2A][0-2] - upcase normal/filename characters
+ DOS-65-23 - YesNo prompt character
+ DOS-65-01 - Get Extended Country Information
+ DOS-65-0[24] - Get pointer to normal/filename upcase table
+ DOS-65-05 - Get pointer to filename terminator table
+ DOS-65-06 - Get pointer to collating sequence table
+ DOS-65-07 - Get pointer to DBCS table
Note: Because I don't know how this used, only an empty table
has been verified to work properly.
+ DOS-66-01 - Get active codepage
+ MUX-14-00 - Installation check
+ MUX-14-02 - Get extended country information
+ MUX-14-04 - Get country information
+ MUX-14-FE - DRDOS get extended country information
+ MUX-14-23 - validate Yes/No prompt (FreeDOS extension)
+ MUX-14-22 - upcase normal character area (FreeDOS extension)
+ MUX-14-A2 - upcase filename character area (FreeDOS extension)
Not implemented is:
+ DOS-65-00 - Change DOS-65-XX information
+ DOS-66-02 - Set active codepage
Note: The rough interface is available, but no code to actually
to change the codepage.
+ external TSR "NLSFUNC"
+ MUX-14-FF - DRDOS prepare codepage
Not validated is:
+ DOS-38 - Set Country code, because:
1) it relies on DOS-66-02 (Set active code page) and
2) requires external NLSFUNC.
+ COUNTRY= statement in CONFIG.SYS (code avilable, but not tested at all)
+ MUX-14-01 - change codepage & MUX-14-03 - Set codepage, because
the meaning of them is not intentional to me. Both function perform
the same request currently, to change the current codepage or country
code or both (though, see DOS-66-02).
= Supported NLS packages
A NLS package may contain data only or data and code.
If the NLS package shall not contain any code, it must conform to the
code already included within the kernel; otherwise an external TSR,
usually NLSFUNC, must provide all code by hooking and intercepting
the MUX-14-XX API.
In order to support the external NLSFUNC, all requests for DOS-XX are
re-routed through MUX-14; but because the NLS API must work, even if no
NLSFUNC has been loaded, the kernel implements a MUX-14 interface of its
own and performs all MUX-14 requests.
However, because the channeling of each request through the MUX chain
is considered a very heavy operation (aka time-consuming), flags are
introduced when to _bypass_ the MUX chain and directly call the
function, which would be activated, if the request would reach the
MUX-14 interface of the kernel.
Because the kernel can only load NLS packages structurally identical to
U.S.A./CP437 per definition, the kernel automatically sets those flags,
thus, retreives all information from them without to channel the request
through the MUX interrupt chain.
The term "structurally identical" is explained in NLS.H.
= Using NLS functions from within the kernel
There are functions to:
+ upcase normal characters: DosUpMem(), DosUpString(), DosUpChar()
+ upcase filename characters: DosUpFMem(), DosUpFString(), DosUpFChar()
+ verify yes/no prompt characters: DosYesNo()
+ retreive data (country informaion): DosGetData() [DOS-65-XX],
DosGetCountryInformation() [DOS-38]
They implement the usual DOS interface and refer to the country code and
codepage by the usual UWORD numbers; NLS_DEFAULT can be used to specify
"current country/codepage". The "Up*()" functions always use the currently
active NLS package.
These functions are also called by the INT-21 handler.
Because of the MUX chain support these functions more or less wrap the
real functions only and check the flags whether to call the internal
function directly or re-route the request through MUX.
Therefore NLS data must not be accessed directly from outside the NLS
implementation, but through these functions only.
CAUTION: The DOS NLS differs between "normal" characters and "filename"
characters, that's why one must call DosUpFString() to upcase a
filename rather than DosUpString()!
Note: The NLS subsystem is robust against any type of characters,
that means DosUpFMem() can be called with any type of junk, except
the pointer to the buffer must not be NULL.
= NLS and fileformats (UNF)
The current implementation does not implemented everything MS-DOS like,
this includes the internal NLS information block and the fileformat of
COUNTRY.SYS. Both structures shall be updated to increase performance,
rather than require the kernel to simulate old and obsolated interfaces.
To overcome the traditional problem with ever-changing structures
a toolset is provided to represent the NLS package in an implementation-
independed way and read/write/manipulate etc. pp. this data.
In the final state NLSFUNC will automatically detect the structures
and transform them into the structure required by the kernel.
To minimize the complexity of these data transformation processes
an independed fileformat called UNF (Uniform NLS file Format) has been
founded, which is totally plain text (except comments) and somewhat
human-readable. Tools will be provided to convert any or particular binary
forms of NLS packages into UNF and back.
Currently available tools:
GRAB_UNF: Extracts all information from the current NLS API and dumps it
into an UNF file. Supports standard information and DOS-65-03 (lowercase).
UNF2HC: Transforms an UNF file into the format of the hardcoded NLS package
ready to be used when the kernel is make'ed.
= Testing / Verifying NLS
Above mentioned UNF toolset includes:
GRAB_UNF: Dump NLS package into UNF file and
NLSUPTST: Test upcase API (DOS-65-2[0-2]).
Testing steps:
1) Generate an UpCase test verifaction file by running "NLSUPTST /c" on
a DOS computer that is entitled to run a good NLS.
Alternatively download an UP file corresponding to your locale,
that means <country>-<codepage>.UP (without the angle brackets).
Note: The numerical country code and codepage must match the settings
of your testee system!
2) Do the same an generate an sample UNF of a good NLS, by running
"GRAB_UNF.EXE" or download one from the internet. the filename is:
<country>-<codepage>.UNF
Note: If you manually edit the file, run "READ_UNF <filename>" to
check the file for errors and dump it in the very same format as
GRAB_UNF will.
3) Copy GRAB_UNF.EXE, NLSUPTST.EXE and the UP file onto the testee, e.g.
floppy. Make sure no .UNF file is located there.
4) Create the CONFIG.SYS with only the minimum settings, more than
a COUNTRY= and a SHELL= are usually NOT required.
5) Create an AUTOEXEC.BAT with this contents (strip leading tabs):
GRAB_UNF.EXE
NLSUPTST.EXE
Note: If you have NLS_DEBUG enabled, a lot of noise will be displayed!
6) Reboot the testee
7) The GRAB_UNF.EXE will display its success by:
"NLS info file for <country>-<codepage> has been created sucessfully"
In this case an UNF file has been created <-> the only way to see this
success status, if NLS_DEBUG is enabled within the kernel.
8) At some point you should see an error message or the good news:
"NLS passed all DOS-65-2[0-2] tests"
This means that NLSUPTST was successful, because there is no other
way to detect this, NLSUPTST must be placed last.
9) Compare the <country>-<codepage>.UNF file form the directory you
run GRAB_UNF.EXE in with the _equally_ named sample file.
Both must be 100% identical, even the number of spaces are.
If COMMAND.COM fails to run AUTOEXEC.BAT, change the SHELL= line within
CONFIG.SYS into:
SHELL=GRAB_UNF.EXE
-and-
SHELL=NLSUPTST.EXE
and boot the testee once with each line.
What do these tests miss?
+ None of these tests try to change neither country code nor code page.
+ By default, these tests cannot override the internal performance flags
and so either the direct-calling or the MUX-re-routing mechanism
is tested, but never both.