darrylo@hpnmdla.HP.COM (Darryl Okahata) (07/12/90)
Here is yet another method for accessing programs in other
directories. It introduces the concept of a Unix-like "PATH" variable,
which gives a list of directories to search. This program, called
"EXEC", works just like Jan Christiaan van Winkel's (jc@atcmp.nl) SVC
function, except that it will search multiple paths. You enter the
arguments to your function on the stack, enter the name of your function
(e.g., 'MYFUNC'), and then execute "EXEC".
To use it, you need a variable called "EXECPATH" in the HOME
directory, and this variable should contain a list of directory paths
(in list form). For example, one possible value for EXECPATH could be:
{ { SYS } { ETC } { ETC ETC } HOME { } }
This causes EXEC to search the following directories for functions:
{ SYS }
{ ETC }
{ ETC ETC }
HOME
the current working directory
Note that HOME does not have to be in a list (for that matter, directory
paths with only one name do not have to be in a list, but it makes it
easier to debug). Also note that placing HOME in this list is
redundant; as everything searches HOME, HOME should never be placed in
this list.
Notes/problems:
1. EXECPATH does not necessarily come from the HOME directory. Strange
things will happen if EXECPATH resides in the "current directory" or
in one of the directories given by EXECPATH.
2. The name "EXECPATH" is too long. It should be shortened to save
time and space.
3. EXEC is slow. The overhead in using EXEC is at least 0.1 sec (I
don't know what the overhead in using SVS is, however). The actual
overhead depends on how large and complex your EXECPATH happens to
be. EXEC should be rewritten in assembly language. It should,
perhaps, keep more things on the stack and not in local variables.
4. EXEC is big (240 bytes).
5. If "{ }" is not contained in the EXECPATH list, the current working
directory will not be searched.
6. Some people will want to remove the code that handles "{ }" (searches
the current working directory). This will increase speed and
decrease program size. To do this, replace the lines that say:
IF DUP SIZE 0
==
THEN DROP cwd
END EVAL
with:
EVAL
7. Searching HOME is redundant.
-- Darryl Okahata
UUCP: {hplabs!, hpcea!, hpfcla!} hpnmd!darrylo
Internet: darrylo%hpnmd@hp-sde.sde.hp.com
DISCLAIMER: this message is the author's personal opinion and does not
constitute the support, opinion or policy of Hewlett-Packard or of the
little green men that have been following him all day.
===============================================================================
Store the following in a variable called "EXEC" in the HOME directory.
Checksum: #19683d
-------------------------------------------------------------------------------
%%HP: T(3)A(D)F(.);
\<< PATH EXECPATH
SIZE 1 \-> cwd end i
\<<
DO EXECPATH i
GET
IF DUP SIZE 0
==
THEN DROP cwd
END EVAL
IF
IFERR RCL 1
THEN
IF 'i'
INCR end >
THEN cwd
EVAL " :NOT FOUND"
+ DOERR
END 0 0
END
THEN cwd EVAL
EVAL 1
END
UNTIL
END
\>>
\>>scottb@hpcvia.CV.HP.COM (Scott_Burke) (07/12/90)
Here is a re-write of EXEC, a program written by Darryl Okahata that implements
path searching on the 48sx. It addresses all of the unresolved issues Darryl
mentions in his posting. It also fixes a bug, is half the size, and preserves
the speed of the original.
Entry syntax: Enter arguments and name of function on the stack. Then execute
'EXEC'. (I assign EXEC to the left-shifted EVAL key, since I never use ->Q.)
The variable 'TRAIL' must be present in the HOME directory. It should be a list
of all "leaves" in the directory tree. To understand this further, imagine a
simple directory structure: (where W1, W2, W3, and GAMES are the "leaves")
HOME
/\
WORK PLAY
/ / \ \
W1 W2 W3 GAMES
The purpose of 'EXEC' is to execute functions contained in the non-current sub-
directory. If a function resides in W2, then 'TRAIL' must contain the list:
{ HOME WORK W2 } for that function to be findable. However, it is only neces-
sary to include the path { HOME WORK } in 'TRAIL' if you SPECIFICALLY want to
search { HOME WORK } BEFORE any of W1, W2, or W3. In that case, put the list
{ HOME WORK } BEFORE { HOME WORK W1 } etc. in the list 'TRAIL'. In the same
vein, it is only necessary to include HOME or { HOME } in 'TRAIL' if it should
be searched BEFORE other subdirectories.
Thus, a normal 'TRAIL' variables for the above structure might be:
{ { HOME WORK W1 } { HOME WORK W2 } { HOME WORK W3 } { HOME PLAY GAMES } }
In Darryl's version of EXEC, he allowed the inclusion of a blank list { } to
represent the current working directory. I have two approaches to that
problem. First (the elegant way): The current subdirectory ought to be one
of the full lists in 'TRAIL', so nothing special needs to be done. Second
(the other way): IF YOU REQUIRE THAT THE CURRENT DIRECTORY BE SEARCHED FIRST,
then include the optional line of code shown in the listing below. That line
checks for the presence of the function in the current directory BEFORE it does
anything else. However, it slows down the code, and makes it bigger, so I
don't include it in my personal version.
Here are solutions/workarounds to some of Darryl's comments:
> 1. EXECPATH does not necessarily come from the HOME directory. Strange
> things will happen if EXECPATH resides in the "current directory" or
> in one of the directories given by EXECPATH.
I require that 'TRAIL' be present in the HOME directory. It's not a big
deal. If you don't like to clutter your HOME directory, I suggest you
implement the { HOME MAIN } work-around mentioned in a previous note,
where HOME contains a whole bunch of garbage, but MAIN is the normal main
working directory and right-shift ' (HOME) is assigned to << MAIN >>.
> 3. EXEC is slow. The overhead in using EXEC is at least 0.1 sec (I
> don't know what the overhead in using SVS is, however). The actual
> overhead depends on how large and complex your EXECPATH happens to
> be. EXEC should be rewritten in assembly language. It should,
> perhaps, keep more things on the stack and not in local variables.
With a large 'TRAIL' list, you're going to see a performance hit, even
in assembly. Local variables vs. stack operations have negligible speed
differences in the case of the type of code we're writing here. It's
only when you're doing things like filtering lists of tens or hundreds of
objects that it's noticeably faster to do it all on the stack. Therefore,
in many cases, it's faster to use local variables because debugging is SO
MUCH SIMPLER.
> 4. EXEC is big (240 bytes).
Make that 107 bytes (or 124.5 with the current directory check first)
Oh yeah, the bug in the other version... If a function calls other routines
it expects to be in the same directory as it is executing from (so much for
grammar), Darryl's version crashes, because it has already returned to the
current working directory BEFORE it executes the function it recalled to the
stack. This is a matter of personal preference, because it occurs to me
that someone may WANT to execute the function in the current directory; it
may create a global variable that they may want to use. It's not too hard
to modify the code to work that way, but mine executes everything in it's
original directory and THEN returns to the current one.
While I'm thinking of possible bugs, let me mention that my code REQUIRES
the LASTARGS flag to be CLEAR (i.e., enabled). To do this, execute -55 CF.
The reason is that when I do a RCL to try to get the function code, if the
global variable is not present, RCL returns the name of the function for
another try. If LASTARGS is SET (i.e., disabled), then the function name
will have to be replicated each loop iteration or stored in a local
variable. I don't change flag -55 because that is usually an impolite thing
to do; I'd store the flags and recall them, but we're talking about way too
much overhead there for such a small function.
Here's the code for EXEC.
<<
PATH -> p store the current working path
<<
[optional line:]
VARS OVER POS { EVAL } IFT check the current directory first
1 TRAIL SIZE FOR i loop from 1 to the size of TRAIL
TRAIL i GET EVAL extract the next path from TRAIL
IFERR RCL try to RCL the function code to stack
THEN if we fail, don't do anything
ELSE EVAL p EVAL KILL END else execute function, reset path, end
NEXT
p EVAL reset path; function wasn't found
>>
>>
Scott Burke
scottb@hpcvia.cv.hp.com or 503-750-3978scottb@hpcvia.CV.HP.COM (Scott_Burke) (07/13/90)
Oops, small typo. In the optional line of the previous program (EXEC), insert a KILL after the EVAL in the list for the IFT. That way, bad things won't happen if you are going to use the option to search the current directory first. Ho hum. Too early in the morn. Scott Burke scottb@hpcvia.cv.hp.com or 503-750-3978