We hope these notes will save you from confusion and lost sleep when writing Perl scripts on VMS. If you find we've missed something you think should appear here, please don't hesitate to drop a line to vmsperl@perl.org.
Most of the complete Perl resides in the shareable image PerlShr.Exe, which provides a core to which the Perl executable image and all Perl extensions are linked. You should place this image in Sys$Share, or define the logical name PerlShr to translate to the full file specification of this image. It should be world readable. (Remember that if a user has execute only access to PerlShr, VMS will treat it as if it were a privileged shareable image, and will therefore require all downstream shareable images to be INSTALLed, etc.)
Finally, Perl.Exe is an executable image containing the main entry point for Perl, as well as some initialization code. It should be placed in a public directory, and made world executable. In order to run Perl with command line arguments, you should define a foreign command to invoke this image.
The portion of the extension provided by the XS code may be connected to the rest of Perl in either of two ways. In the static configuration, the object code for the extension is linked directly into PerlShr.Exe, and is initialized whenever Perl is invoked. In the dynamic configuration, the extension's machine code is placed into a separate shareable image, which is mapped by Perl's DynaLoader when the extension is "use"d or "require"d in your script. This allows you to maintain the extension as a separate entity, at the cost of keeping track of the additional shareable image. Most extensions can be set up as either static or dynamic.
The source code for an extension usually resides in its own directory. At least three files are generally provided: Extshortname.xs (where Extshortname is the portion of the extension's name following the last "::"), containing the XS code, Extshortname.pm, the Perl library module for the extension, and Makefile.PL, a Perl script which uses the "MakeMaker" library modules supplied with Perl to generate a Descrip.MMS file for the extension.
Finally, you'll need to copy the extension's Perl library module to the [.Extname] subdirectory under one of the directories in @INC, where Extname is the name of the extension, with all "::" replaced by "." (e.g. the library module for extension Foo::Bar would be copied to a [.Foo.Bar] subdirectory).
$ perl Makefile.PL ! Create Descrip.MMS
$ mmk ! Build necessary files
$ mmk test ! Run test code, if supplied
$ mmk install ! Install into public Perl tree
N.B. The procedure by which extensions are built and tested creates several levels (at least 4) under the directory in which the extension's source files live. For this reason if you are running a version of VMS prior to V7.1 you shouldn't nest the source directory too deeply in your directory structure lest you exceed RMS' maximum of 8 levels of subdirectory in a filespec. (You can use rooted logical names to get another 8 levels of nesting, if you can't place the files near the top of the physical directory structure.)
VMS support for this process in the current release of Perl is sufficient to handle most extensions. However, it does not yet recognize extra libraries required to build shareable images which are part of an extension, so these must be added to the linker options file for the extension by hand. For instance, if the PGPLOT extension to Perl requires the PGPLOTSHR.EXE shareable image in order to properly link the Perl extension, then the line "PGPLOTSHR/Share" must be added to the linker options file PGPLOT.Opt produced during the build process for the Perl extension.
By default, the shareable image for an extension is placed in the [.lib.site_perl.autoArch.Extname] directory of the installed Perl directory tree (where Arch is VMS_VAX or VMS_AXP, and Extname is the name of the extension, with each "::" translated to "."). (See the MakeMaker documentation for more details on installation options for extensions.) However, it can be manually placed in any of several locations:
If the shareable image isn't in any of these places, you'll need to define a logical name Extshortname, where Extshortname is the portion of the extension's name after the last "::", which translates to the full file specification of the shareable image.
We've tried to minimize the dependence of Perl library modules on Unix syntax, but you may find that some of these, as well as some scripts written for Unix systems, will require that you use Unix syntax, since they will assume that '/' is the directory separator, etc. If you find instances of this in the Perl distribution itself, please let us know, so we can try to work around them.
Also when working on Perl programs on VMS, if you need a syntax in a specific operating system format, then you need either to check the appropriate DECC$ feature logical, or call a conversion routine to force it to that format.
The feature logical name DECC$FILENAME_UNIX_REPORT modifies traditional Perl behavior in the conversion of file specifications from UNIX to VMS format in order to follow the extended character handling rules now expected by the CRTL. Specifically, when this feature is in effect, the "./.../" in a UNIX path is now translated to "[.^.^.^.]" instead of the traditional VMS "[...]". To be compatible with what MakeMaker expects, if a VMS path cannot be translated to a UNIX path, it is passed through unchanged, so "unixify("[...]")" will return "[...]".
The handling of extended characters is largely complete in the VMS-specific C infrastructure of Perl, but more work is still needed to fully support extended syntax filenames in several core modules. In particular, at this writing PathTools has only partial support for directories containing some extended characters.
There are several ambiguous cases where a conversion routine cannot determine whether an input filename is in UNIX format or in VMS format, since now both VMS and UNIX file specifications may have characters in them that could be mistaken for syntax delimiters of the other type. So some pathnames simply cannot be used in a mode that allows either type of pathname to be present. Perl will tend to assume that an ambiguous filename is in UNIX format.
Allowing ``.'' as a version delimiter is simply incompatible with determining whether a pathname is in VMS format or in UNIX format with extended file syntax. There is no way to know whether ``perl-5.8.6'' is a UNIX ``perl-5.8.6'' or a VMS ``perl-5.8;6'' when passing it to unixify() or vmsify().
The DECC$FILENAME_UNIX_REPORT logical name controls how Perl interprets filenames to the extent that Perl uses the CRTL internally for many purposes, and attempts to follow CRTL conventions for reporting filenames. The DECC$FILENAME_UNIX_ONLY feature differs in that it expects all filenames passed to the C run-time to be already in UNIX format. This feature is not yet supported in Perl since Perl uses traditional OpenVMS file specifications internally and in the test harness, and it is not yet clear whether this mode will be useful or useable. The feature logical name DECC$POSIX_COMPLIANT_PATHNAMES is new with the RMS Symbolic Link SDK and included with OpenVMS v8.3, but is not yet supported in Perl.
N. B. It is very easy to get tripped up using a mixture of other programs, external utilities, and Perl scripts that are in varying states of being able to handle case preservation. For example, a file created by an older version of an archive utility or a build utility such as MMK or MMS may generate a filename in all upper case even on an ODS-5 volume. If this filename is later retrieved by a Perl script or module in a case preserving environment, that upper case name may not match the mixed-case or lower-case expections of the Perl code. Your best bet is to follow an all-or-nothing approach to case preservation: either don't use it at all, or make sure your entire toolchain and application environment support and use it.
OpenVMS Alpha v7.3-1 and later and all version of OpenVMS I64 support case sensitivity as a process setting (see "SET PROCESS /CASE_LOOKUP=SENSITIVE"). Perl does not currently suppport case sensitivity on VMS, but it may in the future, so Perl programs should use the "File::Spec-"case_tolerant> method to determine the state, and not the $^O variable.
$ perl -e "print join(' ',@ARGV)" perl.*
perl.c perl.exe perl.h perl.obj
in the following triple quoted manner:
$ perl -e "print join(' ',@ARGV)" """perl.*"""
perl.*
In both the case of unquoted command line arguments or in calls to "glob()" VMS wildcard expansion is performed. (csh-style wildcard expansion is available if you use "File::Glob::glob".) If the wildcard filespec contains a device or directory specification, then the resultant filespecs will also contain a device and directory; otherwise, device and directory information are removed. VMS-style resultant filespecs will contain a full device and directory, while Unix-style resultant filespecs will contain only as much of a directory path as was present in the input filespec. For example, if your default directory is Perl_Root:[000000], the expansion of "[.t]*.*" will yield filespecs like ``perl_root:[t]base.dir'', while the expansion of "t/*/*" will yield filespecs like ``t/base.dir''. (This is done to match the behavior of glob expansion performed by Unix shells.)
Similarly, the resultant filespec will contain the file version only if one was present in the input filespec.
You may also use backticks to invoke a DCL subprocess, whose output is used as the return value of the expression. The string between the backticks is handled as if it were the argument to the "system" operator (see below). In this case, Perl will wait for the subprocess to complete before continuing.
The mailbox (MBX) that perl can create to communicate with a pipe defaults to a buffer size of 512. The default buffer size is adjustable via the logical name PERL_MBX_SIZE provided that the value falls between 128 and the SYSGEN parameter MAXBUF inclusive. For example, to double the MBX size from the default within a Perl program, use "$ENV{'PERL_MBX_SIZE'} = 1024;" and then open and use pipe constructs. An alternative would be to issue the command:
$ Define PERL_MBX_SIZE 1024
before running your wide record pipe program. A larger value may improve performance at the expense of the BYTLM UAF quota.
Perl on VMS supports an emulation of the forked debugger when Perl is run on a VMS system that has X11 support installed.
To use the forked debugger, you need to have the default display set to an X-11 Server and some environment variables set that Unix expects.
The forked debugger requires the environment variable "TERM" to be "xterm", and the environment variable "DISPLAY" to exist. "xterm" must be in lower case.
$define TERM "xterm" $define DISPLAY "hostname:0.0"
Currently the value of "DISPLAY" is ignored. It is recommended that it be set to be the hostname of the display, the server and screen in UNIX notation. In the future the value of DISPLAY may be honored by Perl instead of using the default display.
It may be helpful to always use the forked debugger so that script I/O is separated from debugger I/O. You can force the debugger to be forked by assigning a value to the logical name <PERLDB_PIDS> that is not a process identification number.
$define PERLDB_PIDS XXXX
This allows the programmer to look at the execution stack and variables to find out the cause of the exception. As the debugger is being invoked as the Perl interpreter is about to do a fatal exit, continuing the execution in debug mode is usally not practical.
Starting Perl in the VMS debugger may change the program execution profile in a way that such problems are not reproduced.
The "kill" function can be used to test this functionality from within a program.
In typical VMS style, only the first letter of the value of this logical name is actually checked in a case insensitive mode, and it is considered enabled if it is the value ``T'',``1'' or ``E''.
This logical name must be defined before Perl is started.
In addition, output may be piped to a subprocess, using the character '|'. Anything after this character on the command line is passed to a subprocess for execution; the subprocess takes the output of Perl as its input.
Finally, if the command line ends with '&', the entire command is run in the background as an asynchronous subprocess.
On newer 64 bit versions of OpenVMS, a process setting now controls if the quoting is needed to preserve the case of command line arguments.
file tests*, abs, alarm, atan, backticks*, binmode*, bless,
caller, chdir, chmod, chown, chomp, chop, chr,
close, closedir, cos, crypt*, defined, delete, die, do, dump*,
each, endgrent, endpwent, eof, eval, exec*, exists, exit, exp,
fileno, flock getc, getgrent*, getgrgid*, getgrnam, getlogin, getppid,
getpwent*, getpwnam*, getpwuid*, glob, gmtime*, goto,
grep, hex, ioctl, import, index, int, join, keys, kill*,
last, lc, lcfirst, lchown*, length, link*, local, localtime, log, lstat, m//,
map, mkdir, my, next, no, oct, open, opendir, ord, pack,
pipe, pop, pos, print, printf, push, q//, qq//, qw//,
qx//*, quotemeta, rand, read, readdir, readlink*, redo, ref, rename,
require, reset, return, reverse, rewinddir, rindex,
rmdir, s///, scalar, seek, seekdir, select(internal),
select (system call)*, setgrent, setpwent, shift, sin, sleep,
socketpair, sort, splice, split, sprintf, sqrt, srand, stat,
study, substr, symlink*, sysread, system*, syswrite, tell,
telldir, tie, time, times*, tr///, uc, ucfirst, umask,
undef, unlink*, unpack, untie, unshift, use, utime*,
values, vec, wait, waitpid*, wantarray, warn, write, y///
The following functions were not implemented in the VMS port, and calling them produces a fatal error (usually) or undefined behavior (rarely, we hope):
chroot, dbmclose, dbmopen, fork*, getpgrp, getpriority,
msgctl, msgget, msgsend, msgrcv, semctl,
semget, semop, setpgrp, setpriority, shmctl, shmget,
shmread, shmwrite, syscall
The following functions are available on Perls compiled with Dec C 5.2 or greater and running VMS 7.0 or greater:
truncate
The following functions are available on Perls built on VMS 7.2 or greater:
fcntl (without locking)
The following functions may or may not be implemented, depending on what type of socket support you've built into your copy of Perl:
accept, bind, connect, getpeername,
gethostbyname, getnetbyname, getprotobyname,
getservbyname, gethostbyaddr, getnetbyaddr,
getprotobynumber, getservbyport, gethostent,
getnetent, getprotoent, getservent, sethostent,
setnetent, setprotoent, setservent, endhostent,
endnetent, endprotoent, endservent, getsockname,
getsockopt, listen, recv, select(system call)*,
send, setsockopt, shutdown, socket
The following function is available on Perls built on 64 bit OpenVMS v8.2 with hard links enabled on an ODS-5 formatted build disk. CRTL support is in principle available as of OpenVMS v7.3-1, and better configuration support could detect this.
link
The following functions are available on Perls built on 64 bit OpenVMS v8.2 and later. CRTL support is in principle available as of OpenVMS v7.3-2, and better configuration support could detect this.
getgrgid, getgrnam, getpwnam, getpwuid, setgrent, ttyname
The following functions are available on Perls built on 64 bit OpenVMS v8.2 and later.
statvfs, socketpair
There are DECC feature logical names AND ODS-5 volume attributes that also control what values are returned for the date fields.
Note: Some sites have reported problems when using the file-access tests ("-r", "-w", and "-x") on files accessed via DEC's DFS. Specifically, since DFS does not currently provide access to the extended file header of files on remote volumes, attempts to examine the ACL fail, and the file tests will return false, with $! indicating that the file does not exist. You can use "stat" on these files, since that checks UIC-based protection only, and then manually check the appropriate bits, as defined by your C compiler's stat.h, in the mode value it returns, if you need an approximation of the file's protections.
Note that "binmode" is generally not necessary when using normal filehandles; it is provided so that you can control I/O to existing record-structured files when necessary. You can also use the "vmsfopen" function in the VMS::Stdio extension to gain finer control of I/O to files and devices with different record structures.
The value returned by "crypt" may be compared against the encrypted password from the UAF returned by the "getpw*" functions, in order to authenticate users. If you're going to do this, remember that the encrypted password in the UAF was generated using uppercase username and password strings; you'll have to upcase the arguments to "crypt" to insure that you'll get the proper value:
sub validate_passwd {
my($user,$passwd) = @_;
my($pwdhash);
if ( !($pwdhash = (getpwnam($user))[1]) ||
$pwdhash ne crypt("\U$passwd","\U$name") ) {
intruder_alert($name);
}
return 1;
}
When the future POSIX_EXIT mode is active, "die", the native VMS exit status value will have either one of the $! or $? or $^E or the UNIX value 255 encoded into it in a way that the effective original value can be decoded by other programs written in C, including Perl and the GNV package. As per the normal non-VMS behavior of "die" if either $! or $? are non-zero, one of those values will be encoded into a native VMS status value. If both of the UNIX status values are 0, and the $^E value is set one of ERROR or SEVERE_ERROR severity, then the $^E value will be used as the exit code as is. If none of the above apply, the UNIX value of 255 will be encoded into a native VMS exit status value.
Please note a significant difference in the behavior of "die" in the future POSIX_EXIT mode is that it does not force a VMS SEVERE_ERROR status on exit. The UNIX exit values of 2 through 255 will be encoded in VMS status values with severity levels of SUCCESS. The UNIX exit value of 1 will be encoded in a VMS status value with a severity level of ERROR. This is to be compatible with how the VMS C library encodes these values.
The minimum severity level set by "die" in a future POSIX_EXIT mode may be changed to be ERROR or higher before that mode becomes fully active depending on the results of testing and further review. If this is done, the behavior of c<DIE> in the future POSIX_EXIT will close enough to the default mode that most DCL shell scripts will probably not notice a difference.
See $? for a description of the encoding of the UNIX value to produce a native VMS status containing it.
Also, negative signal values don't do anything special under VMS; they're just converted to the corresponding positive value.
If LIST consists of the empty string, "system" spawns an interactive DCL subprocess, in the same fashion as typing SPAWN at the DCL prompt.
Perl waits for the subprocess to complete before continuing execution in the current process. As described in perlfunc, the return value of "system" is a fake ``status'' which follows POSIX semantics unless the pragma "use vmsish 'status'" is in effect; see the description of $? in this document for more detail.
1 while unlink LIST;
You may need to make this change to scripts written for a Unix system which expect that after a call to "unlink", no files with the names passed to "unlink" will exist. (Note: This can be changed at compile time; if you "use Config" and $Config{'d_unlink_all_versions'} is "define", then "unlink" will delete all versions of a file on the first call.)
"unlink" will delete a file if at all possible, even if it requires changing file protection (though it won't try to change the protection of the parent directory). You can tell whether you've got explicit delete access to a file by using the "VMS::Filespec::candelete" operator. For instance, in order to delete only files to which you have delete access, you could say something like
sub safe_unlink {
my($file,$num);
foreach $file (@_) {
next unless VMS::Filespec::candelete($file);
$num += unlink $file;
}
$num;
}
(or you could just use "VMS::Stdio::remove", if you've installed the VMS::Stdio extension distributed with Perl). If "unlink" has to change the file protection to delete the file, and you interrupt it in midstream, the file may be left intact, but with a changed ACL allowing you delete access.
This behavior of "unlink" is to be compatible with POSIX behavior and not traditional VMS behavior.
Returns PID on success, -1 on error. The FLAGS argument is ignored in all cases.
PERL_ENV_TABLES is translated once when Perl starts up; any changes you make while Perl is running do not affect the behavior of %ENV. If PERL_ENV_TABLES is not defined, then Perl defaults to consulting first the logical name tables specified by LNM$FILE_DEV, and then the CRTL "environ" array.
In all operations on %ENV, the key string is treated as if it were entirely uppercase, regardless of the case actually specified in the Perl expression.
When an element of %ENV is read, the locations to which PERL_ENV_TABLES points are checked in order, and the value obtained from the first successful lookup is returned. If the name of the %ENV element contains a semi-colon, it and any characters after it are removed. These are ignored when the CRTL "environ" array or a CLI symbol table is consulted. However, the name is looked up in a logical name table, the suffix after the semi-colon is treated as the translation index to be used for the lookup. This lets you look up successive values for search list logical names. For instance, if you say
$ Define STORY once,upon,a,time,there,was
$ perl -e "for ($i = 0; $i <= 6; $i++) " -
_$ -e "{ print $ENV{'story;'.$i},' '}"
Perl will print "ONCE UPON A TIME THERE WAS", assuming, of course, that PERL_ENV_TABLES is set up so that the logical name "story" is found, rather than a CLI symbol or CRTL "environ" element with the same name.
When an element of %ENV is set to a defined string, the corresponding definition is made in the location to which the first translation of PERL_ENV_TABLES points. If this causes a logical name to be created, it is defined in supervisor mode. (The same is done if an existing logical name was defined in executive or kernel mode; an existing user or supervisor mode logical name is reset to the new value.) If the value is an empty string, the logical name's translation is defined as a single NUL (ASCII 00) character, since a logical name cannot translate to a zero-length string. (This restriction does not apply to CLI symbols or CRTL "environ" values; they are set to the empty string.) An element of the CRTL "environ" array can be set only if your copy of Perl knows about the CRTL's "setenv()" function. (This is present only in some versions of the DECCRTL; check $Config{d_setenv} to see whether your copy of Perl was built with a CRTL that has this function.)
When an element of %ENV is set to "undef", the element is looked up as if it were being read, and if it is found, it is deleted. (An item ``deleted'' from the CRTL "environ" array is set to the empty string; this can only be done if your copy of Perl knows about the CRTL "setenv()" function.) Using "delete" to remove an element from %ENV has a similar effect, but after the element is deleted, another attempt is made to look up the element, so an inner-mode logical name or a name in another location will replace the logical name just deleted. In either case, only the first value found searching PERL_ENV_TABLES is altered. It is not possible at present to define a search list logical name via %ENV.
The element $ENV{DEFAULT} is special: when read, it returns Perl's current default device and directory, and when set, it resets them, regardless of the definition of PERL_ENV_TABLES. It cannot be cleared or deleted; attempts to do so are silently ignored.
Note that if you want to pass on any elements of the C-local environ array to a subprocess which isn't started by fork/exec, or isn't running a C program, you can ``promote'' them to logical names in the current process, which will then be inherited by all subprocesses, by saying
foreach my $key (qw[C-local keys you want promoted]) {
my $temp = $ENV{$key}; # read from C-local array
$ENV{$key} = $temp; # and define as logical name
}
(You can't just say $ENV{$key} = $ENV{$key}, since the Perl optimizer is smart enough to elide the expression.)
Don't try to clear %ENV by saying "%ENV = ();", it will throw a fatal error. This is equivalent to doing the following from DCL:
DELETE/LOGICAL *
You can imagine how bad things would be if, for example, the SYS$MANAGER or SYS$SYSTEM logical names were deleted.
At present, the first time you iterate over %ENV using "keys", or "values", you will incur a time penalty as all logical names are read, in order to fully populate %ENV. Subsequent iterations will not reread logical names, so they won't be as slow, but they also won't reflect any changes to logical name tables caused by other programs.
You do need to be careful with the logical names representing process-permanent files, such as "SYS$INPUT" and "SYS$OUTPUT". The translations for these logical names are prepended with a two-byte binary value (0x1B 0x00) that needs to be stripped off if you wantto use it. (In previous versions of Perl it wasn't possible to get the values of these logical names, as the null byte acted as an end-of-string marker)
While Perl attempts to keep the vaxc$errno value to be current, if errno is not EVMSERR, it may not be from the current operation.
The next 8 bits contain the termination status of the program.
If the child process follows the convention of C programs compiled with the _POSIX_EXIT macro set, the status value will contain the actual value of 0 to 255 returned by that program on a normal exit.
With the _POSIX_EXIT macro set, the UNIX exit value of zero is represented as a VMS native status of 1, and the UNIX values from 2 to 255 are encoded by the equation:
VMS_status = 0x35a000 + (unix_value * 8) + 1.
And in the special case of unix value 1 the encoding is:
VMS_status = 0x35a000 + 8 + 2 + 0x10000000.
For other termination statuses, the severity portion of the subprocess' exit status is used: if the severity was success or informational, these bits are all 0; if the severity was warning, they contain a value of 1; if the severity was error or fatal error, they contain the actual severity bits, which turns out to be a value of 2 for error and 4 for severe_error. Fatal is another term for the severe_error status.
As a result, $? will always be zero if the subprocess' exit status indicated successful completion, and non-zero if a warning or error occurred or a program compliant with encoding _POSIX_EXIT values was run and set a status.
How can you tell the difference between a non-zero status that is the result of a VMS native error status or an encoded UNIX status? You can not unless you look at the ${^CHILD_ERROR_NATIVE} value. The ${^CHILD_ERROR_NATIVE} value returns the actual VMS status value and check the severity bits. If the severity bits are equal to 1, then if the numeric value for $? is between 2 and 255 or 0, then $? accurately reflects a value passed back from a UNIX application. If $? is 1, and the severity bits indicate a VMS error (2), then $? is from a UNIX application exit value.
In practice, Perl scripts that call programs that return _POSIX_EXIT type status values will be expecting those values, and programs that call traditional VMS programs will either be expecting the previous behavior or just checking for a non-zero status.
And success is always the value 0 in all behaviors.
When the actual VMS termination status of the child is an error, internally the $! value will be set to the closest UNIX errno value to that error so that Perl scripts that test for error messages will see the expected UNIX style error message instead of a VMS message.
Conversely, when setting $? in an END block, an attempt is made to convert the POSIX value into a native status intelligible to the operating system upon exiting Perl. What this boils down to is that setting $? to zero results in the generic success value SS$_NORMAL, and setting $? to a non-zero value results in the generic failure status SS$_ABORT. See also ``exit'' in perlport.
With the future POSIX_EXIT mode set, setting $? will cause the new value to also be encoded into $^E so that the either the original parent or child exit status values of 0 to 255 can be automatically recovered by C programs expecting _POSIX_EXIT behavior. If both a parent and a child exit value are non-zero, then it will be assumed that this is actually a VMS native status value to be passed through. The special value of 0xFFFF is almost a NOOP as it will cause the current native VMS status in the C library to become the current native Perl VMS status, and is handled this way as consequence of it known to not be a valid native VMS status value. It is recommend that only values in range of normal UNIX parent or child status numbers, 0 to 255 are used.
The pragma "use vmsish 'status'" makes $? reflect the actual VMS exit status instead of the default emulation of POSIX status described above. This pragma also disables the conversion of non-zero values to SS$_ABORT when setting $? in an END block (but zero will still be converted to SS$_NORMAL).
Do not use the pragma "use vmsish 'status'" with the future POSIX_EXIT mode, as they are at times requesting conflicting actions and the consequence of ignoring this advice will be undefined to allow future improvements in the POSIX exit handling.