The Weird and The Wonderful

Changing CWD can be a pain in the neck (or maybe even lower down).

Even seasoned programmers may believe that they change CWD, complaining that it fails. That is because the way they do it causes a new process to be spun off, and the CWD applies to that process. When it terminates, "Task completed", the changed WD dies with that process. Make sure that the process executing the CWD is not a child process. If you call some API with a CWD parameter, it might start a new process and use the parameter to set the CWD for that process, without affecting the CWD of the calling process.

You have the same situation with the PATH value: A child process cannot modify the path of its ancestor process. It could change the user profile to change the initial PATH for shells started at a later time, but the shell reads the user profile at startup, and is not affected by profile updates while it is running. The same goes for arbitrarily named environment variables.

Most build systems structure a build in a series of job steps, usually with each job step run as a separate process. Countless developers have tried to modify the environment (CWD, PATH or other environment variables) in one job step, expecting the modification to apply to later steps. It fails, even if the the first step updates the user profile. True enough: The next job step process is started after the profile update, but it is (in several build systems, maybe all) a plain child process, not a shell, so it inherits the environment of the build system process that creates it. The user profile is read by the shell when initializing, not at a plain process creation.

In my last job, we had a large crowd of very qualified system developers. Yet, I had to explain again and again why their CWD, PATH and environment updates "failed". I urged them to forget about inheriting the environment by child processes, and rather transfer such values through the Registry, but two thirds of the SW development group were Linux developers by heart, having been forced to accept working under Windows to get a job, bluntly refusing to use any Windows mechanism unknown in Linux if there was any way to avoid it. As long as they could bitch at lengths about the worthless environment handling in Windows, not realizing that a Linux build system would behave in exactly the same way.

So: When you think that you have set the WD of a given process, check it out to be 100% sure: After modifying it, retrieve the CWD to a variable you can inspect in a debugger, or print it out e.g. on the error device.

It's as good a guess as I've been able to come up with! Thumbs Up | :thumbsup:

I haven't ruled out Cortana intercepting the call and redirecting it to where it thinks (based on the filename?) it needs to go...only to find the callee doesn't live there! As you mentioned, the term 'Mosaic' has been around for a long time in the www lexicon, so I can sorta see a vague connection.

For now, it's a good catch by my tester. I have already adapted (given in) and getting past it. I really only posted here for posterity...on the off-chance that Google may provide it to the next poor sod who can't understand why something (a core win32 method ffs) that has worked for 20 years, suddenly fails on a new Windows version.

Note to my future self: As crazy as it sounds, try changing the filename. Laugh | :laugh:

"Go forth into the source" - Neal Morse
"Hope is contagious"

My freshman year at the Tech. University was the last year of two huge Univac 1100 mainframes as the primary computers for student exercises. Those who have ever worked with the EXEC-8 OS will remember that the file system had "files", which were fairly large disk areas that you organized in "elements" which were e.g. source program units (i.e. what we call "files" in other systems).

Usually, students were given a single such "file" to hold all their work, customary named by the student's name. One of my classmates were refused: Her name was not accepted as a file name, and we could not see the reason.

We asked the system managers, who explained that Exec-8 used a Master File Directory organized as a hash table on the complete file "path" (not a hierarchical one, but a concatenation of a project name and the file name. The code doing the hashing could not handle bucket overflow. So the rejection of my classmate's name really was "Sorry, this hash bucket is full - please select a file name hashing to another bucket!"

Does it only happen on computers with a certain browser installed?

Instead of ShellExecute, can you use a different API like CreateProcess?

If I ever have the time to investigate this further, I can give it a shot, however I found it easier to take the path of least resistance which was renaming the offending executables and getting on with more important items. (I gave in!) Sigh | :sigh:

The whole thing is still a mystery to me. Confused | :confused:

I'll be getting a Win11 development system in the next few months which may shed some light on things. Thanks for the suggestion! Thumbs Up | :thumbsup:

"Go forth into the source" - Neal Morse
"Hope is contagious"

kmoorevs wrote:
Doing so incorrectly starts a web browser with the message 'Hmmm, I can't find that file...Typo?'.

Mosaic is the name of a web browser from the mid 1990s. Years later Mosaic became Netscape and then later became Mozilla Firefox. It would appear that Windows still has a few VERY OLD bits of code lurking around.

In a box in the basement, I came across a short source listing from my student days ...

I was a TA at the time. The 'Programming 101' freshmen had been given a fairly simple programming exercise. We, the TAs, were taking the 'Compiler Technics' course, learning that in FORTRAN, there are no reserved words, only tokens with predefined meaning. The context reveals whether the predefined semantics apply, or if the word is a user defined symbol.

So we made up a self-defined exercize: Solve the 'Programming 101' exercize in FORTRAN, using predefined words only, and no numeric constants. (We found no way to generate character values for the WRITE statements, and had to allow ourselves the use of strings.)

Here is the result. We were using a pre-FORTRAN-77 compiler for the development, getting a nicely formatted output. When the F77 compiler arrived, the output was all left justified, but 'mathematically correct'. We tested it on two different F77 compilers, on two diffent machines, and honestly believe that this code is fully F77 compliant. Unfortunately, I do not currently have access to any F77 compiler, so I cannot verify that (and I really, really hope that it is not Fortran 2018 compliant!)

In FORTRAN, space was insignificant. We had a 'compact' version with all the space removed, which made it rather difficult to follow the program flow. So I give you the readable version. One 🍺 to the first one to tell what the printout from this program will be!

FORTRAN

.
      PROGRAM PROGRAM

      INTEGER IF, INTEGER, GOTO, IMPLICIT
      REAL REAL, DIMENSION, EXTERNAL, FORMAT, END
      INTEGER LOGICAL
      REAL COMPLEX, DATA, CALL, ASSIGN, CHARACTER

      DO FOR IF = INTEGER, INTEGER; END DO
      INTEGER = IF + IF
      GOTO = INTEGER * INTEGER * INTEGER * INTEGER - INTEGER - IF

      CALL FUNCTION(IMPLICIT, REAL, DIMENSION, EXTERNAL, FORMAT, END
     $, LOGICAL, COMPLEX, DATA, CALL, ASSIGN, CHARACTER)

      CALL SUBROUTINE(IMPLICIT, LOGICAL, GOTO, IF, INTEGER)

      END


      SUBROUTINE FUNCTION(IMPLICIT, REAL, DIMENSION, EXTERNAL, FORMAT
     $, END, LOGICAL, COMPLEX, DATA, CALL, ASSIGN, CHARACTER)
      RETURN
      END


      SUBROUTINE SUBROUTINE(IMPLICIT, LOGICAL, GOTO, IF, INTEGER)

      INTEGER GOTO, IMPLICIT(GOTO), LOGICAL(GOTO), IF, INTEGER
     $, EXTERNAL, RETURN

      DO FOR EXTERNAL = IF, GOTO
          DO FOR RETURN = EXTERNAL, EXTERNAL - IF
            IMPLICIT(RETURN) = LOGICAL(RETURN) + LOGICAL(RETURN - IF)
          END DO
          IMPLICIT(IF) = IF
          IMPLICIT(EXTERNAL) = IF
          DO FOR RETURN = IF , GOTO - EXTERNAL
            WRITE(IF, '(''$  '')')
          END DO
          DO FOR RETURN = IF, EXTERNAL
            WRITE(IF, '(''$''I4)') IMPLICIT(RETURN)
          END DO
          WRITE(IF, '( /)')
          DO FOR RETURN = IF, GOTO
            LOGICAL(RETURN) = IMPLICIT(RETURN)
          END DO
      END DO
      END

I have no idea what this code does. However, I did find a free compiler online for you: Free FORTRAN Compilers[^]

Graeme

"I fear not the man who has practiced ten thousand kicks one time, but I fear the man that has practiced one kick ten thousand times!" - Bruce Lee

The last time I looked at a Fortran listing was in school in the fall of 1977. It's been a while!

I instantly thought to myself, why in the hell would anyone on planet earth still use Fortran...

Quick Google search reveals that scientists at NASA and CERN still use Fortran to this day for mission critical work. WTF | :WTF:

Don't forget old guru C.A.R. Hoares's remark after studying the various proposals for what became Fortran-77: "I don't know what programming languages will look like in year 2000, but they will be called 'Fortran'!"

He most definitely was right. My own experience is with Fortran IV and Fortran-77. When I was presented with a Fortran 2008 code sample, I first refused to believe that it was called 'Fortran', bearing very little resemblance to the Fortran I knew. Fortran is very much like 'Ethernet': The ethernet of today has very few similarities with the 3 Mbps linear bus topology thick coax with true CSMA/CD medium access control of my 'computing childhood'. The main similarity, both for Ethernet and Fortran, is the name.

A few years ago a CAD system developer told me they wrote the product in Fortran. In my naivety I said I didn't think anyone still used Fortran. They burst out in hoots of laughter; it turns out a great deal of mathemics-based software is still written in Fortran.

1) is it possible that you added a space after DO FOR... that should not be there?
2) no DATA statements or other initialization of the varables?

Pound to fit, paint to match

Spaces are completely ignored in F77 (and earlier). Columns are significant - specifically col 6. This would be another valid version of the program:

.     PROGRAMPROGRAM
      INTEGERIF,INTEGER,GOTO,IMPLICIT
      REALREAL,DIMENSION,EXTERNAL,FORMAT,END
      INTEGERLOGICAL
      REALCOMPLEX,DATA,CALL,ASSIGN,CHARACTER
      DOFORIF=INTEGER,INTEGER;ENDDO
      INTEGER=IF+IF
      GOTO=INTEGER*INTEGER*INTEGER*INTEGER-INTEGER-IF
      CALLFUNCTION(IMPLICIT,REAL,DIMENSION,EXTERNAL,FORMAT,END
     $,LOGICAL,COMPLEX,DATA,CALL,ASSIGN,CHARACTER)
      CALLSUBROUTINE(IMPLICIT, LOGICAL, GOTO, IF, INTEGER)
      END
      SUBROUTINEFUNCTION(IMPLICIT,REAL,DIMENSION,EXTERNAL,FORMAT
     $,END,LOGICAL,COMPLEX,DATA,CALL,ASSIGN,CHARACTER)
      RETURN
      END
      SUBROUTINESUBROUTINE(IMPLICIT,LOGICAL,GOTO,IF,INTEGER)
      INTEGERGOTO,IMPLICIT(GOTO),LOGICAL(GOTO),IF,INTEGER
     $,EXTERNAL,RETURN
      DOFOREXTERNAL=IF,GOTO
      DOFORRETURN=EXTERNAL,EXTERNAL-IF
      IMPLICIT(RETURN)=LOGICAL(RETURN)+LOGICAL(RETURN-IF)
      ENDDO
      IMPLICIT(IF)=IF
      IMPLICIT(EXTERNAL)=IF
      DOFORRETURN=IF,GOTO-EXTERNAL
      WRITE(IF,'(''$ '')')
      ENDDO
      DOFORRETURN=IF,EXTERNAL
      WRITE(IF,'(''$''I4)')IMPLICIT(RETURN)
      ENDDO
      WRITE(IF,'( /)')
      DOFORRETURN=IF,GOTO
      LOGICAL(RETURN)=IMPLICIT(RETURN)
      ENDDO
      ENDDO
      END

This is (printable) character by character identical to the first version I presented. I am not 100% sure that F77 allows semicolon to put multiple statements on one line; there is one case of it in the above code, and both the F77 compilers we had access to accepted it. I could have filled up the lines to col.72 to make the code even more compact Smile | :)

, but I'll leave that as an exercise for those with extreme tastes in code layout ...

lewist57 wrote:
2) no DATA statements or other initialization of the varables?

We had denied ourselves the use of literals (but had to lift those restrictions for the strings in the WRITE statements).

However: F77 specifies that integer values are initialized to zero. Fortran loops are always executed at least once. So INTEGER starts at 0, and when running

DO FOR IF = INTEGER, INTEGER; ENDDO

the loop will increment INTEGER from 0 to 1. Then we create 2 by adding IF + IF, and 13 by INTEGER*INTEGER*INTEGER*INTEGER-INTEGER-IF (i.e. 2*2*2*2 - 2 - 1).

In principle, we could have created ASCII character codes for the string constants in a COMMMON block, and addressed the block alternately as integer or string. That would require knowledge about how a string is laid out in memory, and the code would have to be split in two files (one file can only interpret the common block in a single way). I believe that Hollerith constants are defined to be a plain octet sequence in memory - no descriptor, no terminator - but Hollerith was left out of F77 (and all later versions), wasn't it?

Today, I have forgotten why we have this no-op subroutine named FUNCTION. It could be for some other strange initialization purpose. All I remember is that it had to be there. If I dig up an F77 compiler and debugger, maybe I could tell. My life does not depend on knowing the answer, so I doubt that I will acquire and install an F77 for the sole purpose of finding out Smile | :)

A couple more things

1) I think only IV/66 always went though a do loop, and 77 and beyond does not
2) I asked about the space between FOR and IF because there is not DO FOR ..... loop. I ran your code through Simply Fotran compiler set for legacy code and it thinks "FOR IF" (or "FORIF") is a variable, which does fit the syntax DO FORIF = variable, variable.

Pound to fit, paint to match

IIRC a satellite was once lost because someone did something like

FORTRAN

DO 10 I = 1. 10

(which is an assignment statement setting a real variable named DO10I to 1.1) rather than

FORTRAN

DO 10 I = 1, 10

(which is a loop from 1 to 10 using an integer called I ending at statement with a label of 10).

I never really liked F77 as it dropped Hollerith constants and ending loops with ENDDO instead of CONTINUE seemed alien. However, I did appreciate the introduction of the block IF statement; especially as I had written and used in production code a pre-processor that supported structured constructs (including IF ... END IF) that generated FIV/66 code - obviously it was written in FIV/66.

jsc42 wrote:
... I had written and used in production code a pre-processor that supported structured constructs (including IF ... END IF) that generated FIV/66 code

Something like Ratfor[^]? I used it back in the day.

Software rusts. Simon Stephenson, ca 1994. So does this signature. me, 2012

Not as sophisticated as that! Mine still used FORTRANesque syntax; in fact, it looked not dissimilar to F77 but that was coincidental as I hadn't come across F77 at the time.

lewist57 wrote:
1) I think only IV/66 always went though a do loop, and 77 and beyond does not

In my basement, I found the manual for one of the two F77 compilers we were using (I never owned a manual for the other), and to my surprise, you are right! The test is done at the top of the loop. I am really surprised that this didn't break thousands of old Fortran programs - changing the semantics of something as fundamental as a DO-loop is rather sensitive!

Anyway: This loop is executed once; it isn't terminated until the DO-variable is larger than the terminal value. "DO I = 0,0; ENDDO" is executed once, and I is incremented from 0 to 1.

2) I asked about the space between FOR and IF because there is not DO FOR ..... loop. I ran your code through Simply Fotran compiler set for legacy code and it thinks "FOR IF" (or "FORIF") is a variable, which does fit the syntax DO FORIF = variable, variable.

jcs42 reminded us of the Mariner crash, where a comma in a loop spec was mistyped as a decimal point, making the DO statement being compiled as an assignment. In the loop in the code I presented, a comma is used, so that DO statement would never be interpreted as an assignment.

I checked the old Fortran manual, which describes the "DO FOR ... ENDDO" with the "FOR" marked as optional. This might have been a syntax proposed for F77, but taken out a short time before the standard was nailed. Starting to use a standard before it has formally passed all formal approvals is quite common, and it took a long time for the F77 proposals to be sorted out and churning it through the mill; it did not happen in 1977! Especially if the standardization process has been plagued by fights between competing viewpoints, you may see lots of last-minute changes - and the F77 process was most definitely a war ground. So maybe both our compilers included the construct, as an extension of F77, because important customers had already started using it. With "FOR" being optional, it can simply be removed; it shouldn't affect the program behavior.

However, the manual identifies the semicolon, for putting several statements on a single line, as an extension. I came across a web page describing new features of Fortran 90, including this, so obviously both our F77 compilers were ahead of their time Smile | :)

Now that you have tried to compile the program: Will you remove the 'FOR's and tell us if the program produces a meaningful output? I suspect that the lines will be left justified (that's what we got with both compilers) - they really should be centered, symmetric around a vertical line. If you know just a little bit of math, and the output is correct, you should be able to recognize it, even when left justified!

Thanks for your response.

The Simply Fortran 3 complier did not complain about the "FOR" between DO and variable, I had never seen that syntax, and just wanted to make sure this was not a misunderstanding on my part. My recollection is that prior to, and including F77, there were dozens of "dialects" of Fortran because folks just wanted to add features in their compilers that were not in the standard, resulting in difficulties in porting the program between various machines and compilers.

I also have the Intel Fortran compiler which is supposed to be compliant with F66, and will try to run it through that when I have the time. All of my work is being done as a console program under Windows 10, using the Simply Fortran 3 program.

I am in the process of reverse engineering the code (manually refactoring to a certain degree), and will be happy to share the revised code when done. I do have a few questions for you, my apologies for asking; they involve subroutine SUBROUTINE:

1) The first line in the subroutine (re)declares the variables. I am a little rusty on subs, but it seems that this would reset them all back to zero? (BTW - I had to explicitly declare the variables in the main program as zero; the Simply Fortran compiler is based upon the GNU fortran compiler, and if you don't initialize the variables, it puts a non-zero value in each uninitialized variable, which messes up your program)
2) The compiler does not like it when you have variables declared as integers in the main program, and then re-declare them as one dimensional arrays within the sub (eg - IMPLICIT and LOGICAL)
3) I am reworking the program as a simple console program running under Windows 10. I don't seem to recognize the following syntax:

WRITE(variable, 'string')

what are these WRITE statements actually doing with the variable and then the string?

Last but not least, it appears that the subroutine FUNCTION does absolutely nothing. I did a listing of the variables before and after calling FUNCTION, and there were no changes. Was this by design?

Overall, this is an excellent program to torture new users of Fortran to drive home the point that even though you could use keywords as variables, you really should not. I was also a TA as a mechanical engineer back in the '70s, and would also do things like this to my students.

Pound to fit, paint to match

lewist57 wrote:
My recollection is that prior to, and including F77, there were dozens of "dialects" of Fortran because folks just wanted to add features in their compilers that were not in the standard, resulting in difficulties in porting the program between various machines and compilers.

That certainly matches my memories. I guess that is why the discussions to create the F77 standard looked more like mad dogs fighting than like academic style discussions Smile | :)

lewist57 wrote:
I do have a few questions for you, my apologies for asking; they involve subroutine SUBROUTINE:

Feel free to ask! I fear that I will have to disappoint you: I haven't looked at this code since my student days, 40+ years ago. The detail justifications for how we did it is mostly forgotten long time ago. And, I never programmed Fortran professionally. We did a few projects at the university. Since then, Fortran has just been an entry in the 'competence list' in my CV (and today, that must be considered a lie).

1) I am not sure about the (re)declaration of the values in SUBROUTINE. The code assumes that variables are initialized to zero - I thought that was defined by the Fortran standard. We had denied ourselves the use of (numeric) constants. If you want to honor this requirement, you may e.g. zero INTEGER by adding as the first statement "INTEGER = INTEGER - INTEGER".

2) I am sorry - I can't explain this one. If the compilers we used did allow such redefinition (the non-array arguments are never used), it would be an excellent way of confusing the code reader. Maybe that is why we did it that way. I don't remember.

3) The two arguments are the logical unit number, IF (= 1) being console output. The second argument is a format string. The first WRITE loop is the one indenting the line properly, which apparently is not valid in F77, as both our F77 compilers failed to do proper indentation. Maybe the format string could be updated to F77 standard; it just specifies two space characters to be output. The format is that of the FORMAT statement which requires the format string to be enclosed in parentheses, so the '(' and ')' are just to fulfill this requirement. It seems like the '$' is a quoting, to prevent the tokenizer from stripping away the space.

For the second WRITE loop, it seems to me that the format string could have been written '(I4)'; I have no recollection of why the '$' is there.

I have a vague memory of FUNCTION being there for a purpose, the program not working without it. I may be wrong - maybe we put it in there just to confuse people to believe that it had a purpose Smile | :)

I really appreciate that you spend time trying to make this program work! Unfortunately, I am too far away from you to give you - in the hand, in the meat world - the 🍺 I told I would give to the first one telling what the output of this program is, but you seem to be one who is likely to win the prize. If you decide to spend a vacation in Norway, I'll gladly buy you one!

lewist57 wrote:
Overall, this is an excellent program to torture new users of Fortran to drive home the point that even though you could use keywords as variables, you really should not.

And then ... Have you noticed my entries in the thread 'General Programming | Algorithms | Aide pour un programme langage c'? Through a lengthy discussion, in particular with jschell, I argue in favor of a programming language being defined in terms of a parser tree, with no (reserved or just predefined) keywords. An IDE working directly on a parse tree would represent structural information in binary format, making it visible to the user by e.g. typographic highlighting, but requiring the user to create structure by function keys or menu selections, rather than by keywords. Just like a document processor - it has no (reserved) keywords that you cannot use freely in your document, and there is nothing you can put into the text body that will confuse the binary structure information.

I really wish we could have a language defined by its abstract syntax rather than by textual keywords. But I am not holding my breath waiting for it to appear.

Some more progress:

1) the darn dollar sign $

From your response: "The first WRITE loop is the one indenting the line properly, which apparently is not valid in F77, as both our F77 compilers failed to do proper indentation. Maybe the format string could be updated to F77 standard; it just specifies two space characters to be output."

So my interpretation is that WRITE(IF, '(''$ '')') simply writes two spaces, and can be replaced by write(*,*)' ' where there are two spaces between the single quotes. That makes a lot of sense. The $ sign has been used in various compilers to indicate a macro, or as a suppression of line return, so that helps a lot. It might be some compiler specific feature.

2) The compiler is not happy with the first do loop in SUBROUTINE:

DO FOR RETURN = EXTERNAL, EXTERNAL - IF
IMPLICIT(RETURN) = LOGICAL(RETURN) + LOGICAL(RETURN - IF)
END DO

The complier will not throw an error or warning (either when compiling or running), but as I see it, the first time through EXTERNAL = 0 and IF = 1, so you are doing a negative increment

DO FOR RETURN = 0, -1

when I test a program with a do loop like above, there are no errors or warnings from the compiler, and the program runs, but it apparently just skips over the do loop and its contents.

For example, this code

write(*,*)'before do loop'
do RETURN = 0, -1
write(*,*)'inside do loop'
end do
write(*,*)'after do loop'

produces the following output:
before do loop
after do loop

3) The declaration of variables in the main program have IMPLICIT and LOGICAL as scalars, and then are redeclared as arrays in SUBROUTINE; there is no definition of the size of the array (should be 13). Likewise, EXTERNAL is declared as an real in the main program, and redeclared as an integer in SUBROUTINE.

4) last but not least, F66 and F77 limited variables, subroutines, functions, program names to 6 characters; how did your compiler accept the ones that are 7 or more characters long?

Thanks again!

Pound to fit, paint to match

I just spent two hours debugging a problem with my handheld NES emulator (ESP32 based) not loading a bunch of the ROMs that I have.

When I read certain files I was getting nothing but zeroes back. The file was the right size but it's like all the data was zeroed.

I was pulling my hair out because the same copy of the file on my PC read fine. The one on the SD card I copied from the PC did not.

Turns out the copy was bad somehow. Nothing wrong with my code. However, it seems like (I think, but I need to test more) that if I copy them with long names intact they don't work. If I rename them to a shorter name first, they work. Either way, everything looks like it works until you open the file.

To err is human. Fortune favors the monsters.

So the doohickey uses the old Windows 95 file scheme? Where your could "name" your file something long but in reality it made an 8.3 name for the file in the background?

I’ve given up trying to be calm. However, I am open to feeling slightly less agitated.