Issues in porting libxml to VMS
===============================

Here's a summary of the issues I encountered when building LIBXML under
VMS.  There was some VMS support in the version I got, but it was a little
out of date with the result that some newer files had problems.

I present this list "as is" to hopefully act as a guide to anyone having
problems in the future.

That's it.  Good luck!

John A Fotheringham (jaf jafsoft com)
October 2001


Installation kit
----------------

- File attributes.  Having downloaded essentially a Unix distribution, some
  of the file attributes weren't correct... especially those in the [.VMS]
  subdirectory.  In EDT you could see line feeds and carriage returns as
  <LF><CR> etc.  To correct this use the command

	$ set file <filespec> /attr=rfm=stm

  This sets the record format to be "stream".  Other variants may be used
  instead depending on how you got the files onto your system.  Files will 
  look okay in an EDT editor once the attributes are set.  Without
  this the command file may not run correctly, since it may be interpreted
  as a single line.

- VMS-specific files are in a [.VMS] directory.  If you've found this file
  then you already know this :-)  This directory contains

      BUILD_LIBXML.COM	- a build command file, which I've radically re-worked
      CONFIG.VMS	- a configuration file to replace config.h
      README.VMS        - these readme notes

- Don't execute BUILD_LIBXML.COM until you've done all the following

  - read these notes
  - defined a logical XMLOLB to point to the directory you want object 
    libraries placed in
  - copied CONFIG.VMS to CONFIG.H in the main sourec directory (there
    may be a file CONFIG.H_IN already there)
  - reviewed the configuration section of BUILD_LIBXML.COM, and in particular
    updated the module lists in line with MAKEFILE
  - identified the location of the include files, so that you can manually
    set the LIBXML logical if need be (lookk for globals.h or xmlversion.h)
  - re-read these notes :-p

  instructions for all these steps are below.

- the file [.vms]config.vms should be used to replace config.h.  This file
  contains a number of define statements that identify the software options
  required under VMS

- The include files are in a [.INCLUDE.LIBXML] subdirectory.  You need
  a logical "libxml" to point to this so that include statements of the
  form

	#include <libxml/parser.h>

  will work correctly.  The source files are mostly two levels above this
  directory, although there are some .h files there as well.

- The command file BUILD_LIBXML.COM will do the following

  - setup some logicals
  - set def to the source directory
  - compile modules and place them into a LIBXML.OLB library
  - compile and link a number of self-test programs
  - compile and link a number of utilities and test programs
  - set def back to the original directory (i.e. if it fails you might not be
    where you started :-)

  before running this command file review the configuration segment at
  the top.  In particular compare the lists of modules with those in the
  most recent version of the Unix MAKEFILE.  Instructions are contained
  in the command file itself.

  The command file will attempt to set two logicals

  - xml_srcdir.  The directory containing the source files
  - libxml.  The include file directory.

  It attempts this by looking for modules globals.c and globals.h in
  nearby directories.  If this logic fails, you'll need to manually define
  these logicals.


The TRIO package
----------------
- A sub-package TRIO is used to provide some functions not naturally available
  under VMS.  These include support for infinite and undefined numbers,
  and specialised print functions like "snprintf"

  I had to make several changes to trionan.c in discussion with the author
  (hopefully these are now included in the distro, so I won't list them here)

  To build this software we need to add

	/IEEE_MODE=UNDERFLOW_TO_ZERO/FLOAT=IEEE 

  to the compile command for xpath.c and trio.c, and to any main program
  that uses this functionality.  BUILD_LIBXML.COM should do this for you.


Compiler and linker errors
--------------------------
- the DEC C compiler will produce a number of warnings when compiling the
  C code.  These include

    - Implicit function warnings.  These indicate functions whose type is
      not defined in a .h file.  This will probably only happen if your
      configuration is not correct (e.g. for "snprintf" if you haven't
      edited xmlversion.h to set WITH_TRIO

    - uninitialised variables.  Not usually a problem.  You can solve this
      by editing the code to initialise the variables affected

  and with respect to the trionan code

    - no main module for UNDERFLOW_TO_ZERO
      - in XPATH.C, maybe others
    - float overflow when compiling trionan.c STANDALONE

  You can suppress these "expected" messages by using the compile command

    $ CC /WARN=DISABLE=(FLOATOVERFL,NOMAINUFLO)

  This can be achieved by specifying "NOWARN" as the first command
  parameter to BUILD_LIBXML as follows

    $ @BUILD_LIBXML NOWARN

- the Linker will report the following error

      %LINK-W-MULDEF, symbol DECC$STRERROR multiply defined
        in module DECC$SHR file SYS$COMMON:[SYSLIB]DECC$SHR.EXE;5

  This is complaining that DECC$STRERROR is multiply defined, which in turn 
  is because this system symbol is getting added to LIBXML.OLB when strio.c
  is compiled.

  I'm not sure what the solution for this is, but this is a fairly benign error.


Changes made to the codebase
----------------------------
- I changed all dummy declarations in trio.c to be 

    va_list dummy = NULL;

  to prevent compiler whinge in TRIO.C about uninitialised variables

- I had to add the following to nanoftp.c

    #if defined(VMS) || defined(__VMS)
      #define SOCKLEN_T unsigned int
    #endif

  This matches similar lines already added to nanohttp.c

- Several variables and function names exceed the 31 character limit 
  used in VMS.  This leads to compiler warnings.  The solution adopted
  has been

  a) where variables or functions are defined, set up an ifdef on VMS 
     to define a variable/function of a shorter name.  This may need to
     be done in the .h file, and in the implementation .c file

  b) use define statements to map the long name onto a shorter name.  That
     way all future code can refer to the function/variable by the long name,
     reducing the need to add VMS-specific code everywhere.

  In the current distro, I had to do this for the following names

  in globals.c, parserinternals.c, globals.h, parser.h
    xmlSubstituteEntitiesDefaultValue
    xmlDoValidityCheckingDefaultValue

  in globals.c, globals.h
    __xmlDoValidityCheckingDefaultValue

  in xmlio.c, xmlio.h
    xmlRegisterDefaultInputCallbacks
    xmlRegisterDefaultOutputCallbacks

  in xpath.c and xpathinternals.h
    xmlXPathRegisteredVariablesCleanup