1D Array to String not compiling correctly

[G-CODE]

I was experiencing a whole host of problems running an application in LabVIEW 2020 (32-bit) dev environment. This function was giving me the wrong output. After an edit and save, the output of the function now appears correct. This is the first time I have ever seen this...

 

 

[hooovahh]

Oh man that looks like a painful thing to debug and isolate.  And an even bigger pain to convince NI that there is an issue when someone in the support call will just have a goal of getting your code working.  I haven't seen anything in 2020 like this yet.  I have seen code that application builder says is broken, I go and run the VI it isn't broken, and then the build can continue. It is related in that it seems to be a compiler issue.

[X___]

NI recommends to force recompile (mass compile) after each update.

[G-CODE]

This happened on a virtual machine that only had LabVIEW 2020 installed. VIPM was always configured to mass compile VIs after package installation. If this happened even when packages were being mass compiled, what else is lurking under the hood waiting to be found?

I want to assume that had I built an EXE in this state, this problem would have gone away.

[X___]

Gotcha. That is problematic.

[Antoine Chalons]

out of the blkue, this issue just appeard on my LV2019 after installing LV2020 😮

 

Thx for publishing the fix.

[Jim Kring]

Good news! There's a new release available on vipm.io with a fix.

https://www.vipm.io/package/oglib_string/#4.1.1.16
 

The issue

We were able to figure out what the issue is. It appears to be caused when using a NI Linux RT target.

The problem seems to happen when LabVIEW compiles/saves 1D Array to String for running on an NI Linux RT target. Then, when you open and run 1D Array to String on Windows, it isn't always getting recompiled correctly.
 

More Details

The current implementation of 1D Array to String uses an EOL constant () to build a regular expression -- the EOL is platform dependent (it's CRLF "\r\n" on Windows and LF "\n" on Linux and Mac).

My guess is that somehow the LabVIEW compiler constant folded the regular expression that is the concatenation of the EOL constant and "$" into "\n$" (match a Line Feed at the end of the string). 

However, on Windows the EOL character we are looking for is a CRLF ("\r\n") at the end of the string.  So, if we were searching for "\n$" at the end, the before string would have a "\r" at the end of it, which is what we're seeing (when the bug is happening).

Editing and re-saving this VI on Windows causes it to be recompiled and the regular expression get's correctly compiled and constant folded into "\r\n$"

A More Robust Implementation (work-around)

Looking at that 1D Array to String code, I don't really love how it's building the regular expression using an EOL constant.

I think that a more robust solution would just build the regular expression as "[\r\n]+$" (one or more Carriage Return or Line Feed characters at the end of the string).

This would be platform independent and not suffer from the compiler bug we're discussing here.

There's a new release available on vipm.io with this fix.

https://www.vipm.io/package/oglib_string/#4.1.1.16

[Update] I've opened a service request with NI to dig into it some more. I'll post updates, once I find out more.

Edited February 4, 2021 by Jim Kring

[Rolf Kalbermatter]

17 hours ago, Jim Kring said:

 

The current implementation of 1D Array to String uses an EOL constant () to build a regular expression -- the EOL is platform dependent (it's CRLF "\r\n" on Windows and LF "\n" on Linux and Mac).

LabVIEW for Mac OS Classic used actually a "\r" character for this, since the old MacOS for some reason was using yet another EOL indicator.

Quote

A More Robust Implementation (work-around)

Looking at that 1D Array to String code, I don't really love how it's building the regular expression using an EOL constant.

I think that a more robust solution would just build the regular expression as "[\r\n]+$" (one or more Carriage Return or Line Feed characters at the end of the string).

This would be platform independent and not suffer from the compiler bug we're discussing here.

This is strictly speaking not equivalent to the originally intended implementation since this would remove multiple empty lines at the end while the original one only removed one EOL. However the Array to Spreadsheet String function should not produce empty lines (except maybe if you enter a string array of n * 0 elements) but that does not apply for this 1D array use case.

[Jim Kring]

2 hours ago, Rolf Kalbermatter said:

This is strictly speaking not equivalent to the originally intended implementation since this would remove multiple empty lines at the end while the original one only removed one EOL. However the Array to Spreadsheet String function should not produce empty lines (except maybe if you enter a string array of n * 0 elements) but that does not apply for this 1D array use case.

You're right, Rolf. I can see some situations where my proposed "fix" would fail -- if the last element(s) of the Array of Strings intentionally contains EOL characters at the end  ["\t", "\r", "\n"]

Now, looking at the "1D String Array to Delimited String.vi" that got added to LabVIEW (I'm not sure which version), we see it has a similar implementation (although it doesn't use Match Pattern and simply removes the calculated length of the EOL at the end of the string).

I wonder if this VI suffers from the same constant folding compiler bug reported in this thread (I'm going to do some testing).

 

This all makes me wonder if we should be creating the delimited string in a more low-level way (instead of relying on the Array to Spreadsheet String function, which adds the EOL at the end).

Here's a simple solution:

 

And, here's a potentially more performant solution for larger strings:

Thoughts about the best approach here?

 

[Jim Kring]

Here's another potential "fix" that simply aims to work around the compiler bug and doesn't deviate from the original implementation (which helps avoid any unintended changes in behavior that might be caused by a new implementation).

[Jim Kring]

We've figured out more information.

The problem seems to go away if you turn ON "Separate compiled code from source file" for 1D Array to String

Meaning: the problem occurs when the compiled code is saved in the VI source file.

It would seam that LabVIEW is not correctly determining when it needs to recompile/save the VI when switching between RT and Windows.

[Jim Kring]

OK, here's a game plan I think will work to fix the issue, yet without changing the behavior of the code.

Since LV2009 (the version of LabVIEW the package is currently developed in) doesn't have the "Separate compiled code" feature), we'll implement a work-around by creating a special subVI to get the EOL that is not affected by the bug (Note: ignore the fact that the screenshot shows this VI open in LV2020).

We'll do this by calling the Array to Spreadsheet string function with empty string inputs. It appears from testing that this does not get constant folded (or at least is not impacted by the bug). And, we'll cache this value for performance on subsequent calls. It'll all be packaged into a tidy little sub-VI, so that it's obvious what we're doing and why.

I've released this as version 4.1.2.18.

Moving forward, we'll upgrade the sources for this package to LV2013 and mark all the VIs for "Separate from compiled code".  We would have done that sooner, but there was a build issue in 2013, due to the fact that NI introduced a VI called "Trim Whitespace.vi" that shares the same filename and thus collides with the OpenG VI (in source code form before it's name mangled in the build process). But, that's a different story (and I think we can work around it during the build process)...

[Jim Kring]

This bug has been submitted to NI and confirmed.

bug number 1289111

[rtollert]

We think we got 1289111 licked in LabVIEW 2021.