How about having a timeout occurrence as an input for functions which support timeouts?
I am illustrating a single use case with queues (and a notifier) but I would see this as being beneficial to nearly ALL functions with timeout inputs.
Sometimes we'd like to wait for one of a few different functions (an example of mine which springs to mind is the Dequeue primitive). At the moment we must essentially poll each primitive with a relatively short timeout, see if one responded and then handle accordingly. This is (for me at least) ugly to look at and introduces polling which I generally don't like. I'm an events man.
What I propose would be that instead of simply defining a timeout in milliseconds we can define both a timeout (in milliseconds AND an occurrence for the timeout). If we wire this data to several primitives (all sharing the same occurrence) the first primitive to receive data triggers the occurrence so that the others waiting in parallel return also.
In the case where no data arrives, each function waits the defined amount of time but upon timeout DOES NOT fire the occurrence. This would cover corner cases where you may want different parallel processes to have different timeouts (Yes there are cases although they may be rare). It is possible to change the "priorities" of the incoming queues in thie way.
Background info: One example where I could use this is for RT communication. Here we multiplex many different commands over TCP or UDP. On the API side it would be beneficial to be able to work with several strictly-typed queues to inject data into this communication pipe while still maintining maximum throughput. I don't like using variants or flattened strings to achieve this multiplexing.
Being forced to poll means the code must decide between efficiency (low CPU usage) OR response time (setting a very low timeout). Although the CPU load for polling may be low, it's not something I personally like implementing and I think it's generally to be avoided.
There IS a LVOOP solution to this particular problem but not everyone can use LVOOP in their projects (for various reasons). I can envisage other use cases where interrupting a timeout would be desireable (Event structure, wait on ms, VISA read and so on and so forth).
Currently if a VI is set to subroutine priority you can only call subVIs within that which are also subroutine priority (to prevent priority inversion I guess).
It would be great if it was possible to also use inlined subVIs inside subroutine VIs.
As inlining basically defeats a VI's priority setting an inlined subVI would just "inherit" the subroutine priority of its caller. I configure many of my very small reuse VIs as inlined (most of those in the GPower Error & Warning toolset from v2.1 onwards for instance), since they typically perform much better than subroutine that way. But since they are configured as inlined, this effectively prevents them from being (re)used inside subroutine VIs.
It is very common to build an array or string in a loop using the Build Array or Concatenate String primitives. Unfortunately, while easy, this also incurs a large performance degradation (see this post for some examples) and this is not communicated to the creator in any meaningful way, other than slow code. I propose changing the primitive appearance in some obvious way to show that a performance degradation is happening. This would only happen if the primitive was in a loop, since these functions are a normal part of LabVIEW programming. Some possible methods of showing issues:
Accompanying any of these would be a mouse-over or pop-up to explain why the colors changed. The change would also be explained in the context and primary help. I would definitely want a switch, probably in Tools->Options and the labview.ini file to turn the behavior off for advanced users. New users would see the change and should be able to find out why it is occurring fairly quickly. The help files for the nodes should give alternatives for using them in a loop, such as the post above or replacing a Build Array with a conditional index.
The majority of the problem comes from Build Array, Delete from Array, Insert Into Array, and Concatenate Strings, so an initial pass at this problem could target them only. The Build Array and Concatenate Strings issues could be largely removed by using compiler optimizations similar to the ones currently being used for the conditional index, although algorithm changes by the developer can lead to higher performance than the generic case (see link above). The Delete from Array and Insert Into Array issues are usually solved by algorithm changes on the part of the developer, so an indication of an issue would be useful.
Auto-indexing of arrays in for and while loops are a nice luxury in LabView. One option that could save much time would be a menu option to turn on conditional indexing, this would expose a boolean terminal under the auto-index icon to select if the current itteration should add the itteration to the array or skip it. From an execution standpoint there would only be a minor performance hit (could still preallocate max array size on for loops and automatically return used subset). This could also work for autoindexed in but would have less use that the autoindeded out case. I know I have built many conditional arrays inside of a for loop and it requires a case selection and a build array making the code less readable and requires time and thought. It can also be less efficient than a compiler can do.
See the example below which would run a for loop and only build array of < 0.1
We're witnessing more and more requests to stop LV hiding important information from us. In One direction we want to be able to know (and some want to break code) if structures are hiding code.
Others want LV primitives to give visual feedback as to how they are configured, especially if that configuration can have an effect on what's being executed or how it's executed.
Examples include (Please please feel free to add more in the comments below)
Array to cluster (Cluster size hidden)
Boolean array to number (Sign mode hidden)
FXP simple Math (Rounding, saturation and output type hidden)
SubVI node setup (When right lcicking the subVI on the BD and changing it's properties - show FP when run, suspend and so on)
Sub VI settings in general (Subroutine, debugging)
I know there are already ideas out there for most of these (and I simply chose examples to link to here - I don't mean to leave anyone's ideas out on purpose) but I feel that instead of targetting the individual neurangic points where we have problems, I would like to acknowledge for NI R&D that the idea behind most of these problems (Some of them go much further than simply not hiding the information, and I have given most kudos for that) is that hiding information from us regarding important differences in code execution is a bad thing. I don't mean to claim anyone's thunder. I only decided to post this because of the apparent large number of ideas which have this basic idea at heart. While many of those go further and want additional action taken (Most of which are good and should be implemented) I feel the underlying idea should not be ignored, even if all of the otherwise proposed changes are deemed unsuitable.
My idea can be boiled down to the fact that ALL execution relevant information which is directly applicable to the BD on view should be also VISIBLE on the BD.
As a disclaimer, I deem factors such as FIFO size and Queue size to be extraneous factors which can be externally influenced and thus does not belong under this idea.
Example: I have some Oscilliscope code running on FPGA and had the weirdest of problems where communications worked fine up to (but not including 524288 - 2^19) data points. As it turns out, a single "Boolean array to number" was set to convert the sign of the input number which turned out to be completely wrong. Don't know where that came from, maybe I copied the primitive when writing the code and forgot to set it correctly. My point is that it took me upwards of half a day to track down this problem due to the sheer number of possible error sources I have in my code (It's really complicated stuff in total) and having NO VISUAL CLUE as to what was wrong. Had there been SOME kind of visual clue as to the configuration of this node I would have found the problem much earlier and would be a more productive programmer. Should I have set the properties when writing the code initially, sure but as LV projects grow in complexity these kinds of things are getting to be very burdensome.
I suggest an option added to the Open VI Reference primitive to open that VI reference without any refees. I suggest option bit 10, i.e. option 0x200h:
The demand for this arises when you want to access methods and properties of VIs that may be broken, while on the other hand you don't have any need to run those VIs - for instance in one of my current tools (BatchEditor) where I'm tasked with investigating hundreds of VIs simultaneously of which some could be broken or missing dependencies. Other situations would be tools for traversing project trees for instance. Opening a large number of healthy VI references takes a while, and when something is broken in a VI, opening even a single reference could take 30 seconds to minutes.
Currently you can suppress the "loading" and "searching" dialogs by setting option 0x20h on the Open VI Reference primitive, but that only loads the refnum silently as far as that will get you. Opening the refnum takes the same amount of time as if you could see those dialogs, and you are rewarded with an explorer window if a dependency search fails. I just want a way to open a VI refnum without even starting to look for dependencies, thus very quickly and guaranteed silently.
The relevant people would know that this request isn't that hard to implement, as it is kind of already possibly using some ninja tricks. I'd like such an option to be public.
When creating an installer for my built LabVIEW application, I really dislike having to choose between including the RTE installer (and having a 100+ MB installer for my application) or not including it (and requiring my users to download and install the RTE as a separate step). Typically, I'll build two installers at the same time (with roughly duplicate build settings): a full installer w/ RTE and a light installer w/out the RTE.
What would be much nicer would be if my app's installer were able to download and install the RTE, if necesary. Actually, this is common practice, these days, for users to download a small installer that then downloads larger installer files behind the scenes.
I am extensively using the Array to Spreadsheet string Primitive and most of the time I never used the Format string input (Use to wire an empty string constant) and still I am getting the right result what I expect. So I think it would be better if the Required Terminal is changed to an Optional Terminal.
It is known that the Array to Spreadsheet string is Polymorphic, but when we wire an array of I32 and DBL the output string is DBL format only. It would be good if the output String adapts the data type that is wired unless otherwise specified by the Format string.
The Desktop Execution Trace Toolkit (DETT) can trace enqueue/dequeue operations, so I would think this is feasible:
Add a mechanism to like the Profile>Performance & Memory view to display all active queues in memory by name (for "unnamed" ones, use whatever unique refnum/identifier is available) as well as the max size and current # of items in queue. You could use the same "Snapshot" functionality as the Profiler tool.
A particular use case:
We were tracking a memory leak in a large application that resulted from an unbounded queue whose consumer was disabled. The standard DETT and Profile tools weren't showing where the excess memory consumption was coming from, since queue data does not "belong" to a single VI. Granted, you can see the individual enqueue/dequeue operations in DETT, and even highlight pairs, but that's a little cumbersome in a large application.
There is something wrong with this VI, although you wouldn't know it unless you ran it (and I should warn you that it will annoy you if you run it):
What's wrong with it is that auto grow has been disabled and there's some annoying code hidden there beyond the loop boundary. This is one of my least favorite things about LV - it allows us to hide code completely and get away with it. I don't think it should.
LV already has auto grow enabled by default to handle some of the cases which cause this issue, but I know that many many people don't like it when LV automatically plays with their BD layout (and rightly so) and auto grow only covers some of the cases, so I think we need something more encompassing and less obtrusive, and I would suggest breaking the VI if it has hidden code.
I also know that LV has warnings and VI Analyzer has tests for this, but I think this needs to be something which doesn't let you get away with it.
I think LV should break any VI which has any of the following:
The error ring is a handy tool I've just learned about. It would be fantastic to be able to create an error ring for a project that can be type defined to easily distribute with applications. This way all custom error codes for a project may be easily stored with all the project files without having to worry about including a file from the user.lib. Other ring controls/constants/indicators can be type defined, why can't the error ring?
Format into text is very useful but can become hard to edit when it has a lot of inputs. I propose, instead of one huge format string, that the programmer be allowed to put the required format next to the corresponding input. Also, the user should be allowed to enter constant strings, e.g.. \n, \t, or "Comment", and have the corresponding input field automatically grayed out.
I use Unflatten from string (UFS) a lot. Most of the the flattened data, however, comes from outside LV and as such does not use an I32 to encode the length of strings and arrays. That leads to a lot of branching and bending of wires.
I would like to make the boolean input (Includes length?) polymorphic to accept numeric types to specify the number of elements in an array or the length of a string. I was too lazy to draw it, but I would also like it to promote the type input to an array if the count is wired so I only need a simple scalar constant to specify the type instead of the array constant (think Read from Binary File without the cluster of array business). Allowing a cluster of N numerics to specify ND arrays would be very welcome and permit compile time error checking. 1D arrays for dimension sizes could work as well, but the error checking would be runtime only.
I desperately want the boolean input to be overloaded since it is a straight shot from the output of the previous UFS.
It is time to put a dent in the floating point "problems" encountered by many in LV. Due to the (not so?) well-known limitations of floating point representations, comparisons can often lead to surprising results. I propose a new configuration for the comparison functions when floats are involved, call it "Compare Floats" or otherwise. When selected, I suggest that Equals? becomes "Almost Equal?" and the icon changes to the approximately equal sign. EqualToZero could be AlmostEqualToZero, again with appropriate icon changes. GreaterThanorAlmostEqual, etc.
I do not think these need to be new functions on the palette, just a configuration option (Comparison Mode). They should expose a couple of terminals for options so we can control what close means (# of sig figs, # digits, absolute difference, etc.) with reasonable defaults so most cases we do not have to worry about it. We get all of the ease and polymorphism that comes with the built-in functions.
There are many ways to do this, I won't be so bold as to specify which way to go. I am confident that any reasonable method would be a vast improvement over the current method which is hope that you are never bitten by Equals?.
You don't need to look too deep to find examples of TDMS API "memory leaks". These are not leaks but rather the TDMS API holding meta-data in memory related to the structure of the TDMS file in case you need to read from the file.
The biggest issues with this is that it makes TDMS a bad choice for RT applications where you make cause this index to build through file fragmentation however in most RT applications you are only writing. On investigating whether it would be possible to make an RT compatible API (in native LV) it appears that there is no reason for this to increase in a write only mode.
So the request is a write only API or flag in the existing API. This should prevent memory build up in loggers.
There is a design decision around whether to use a seperate API or have errors thrown if you attempt to read, I think I would be open to either but I think the error would work best to allow easy upgrade of existing applications.
I've done this like a milion times! It's not very "pure", but how convenient would it be if the event structure itself kept looping until stopped?
This should be optional, like in the for loop ("Conditional Terminal"). For the event stucture, it wwhould be nice if wiring it is optional (unwired means keep looping), or that you can show\hide it on a per event basis.
I propose that Case Selectors should accept any type of reference, and the two cases generated are "Valid Ref" and "Invalid Ref". (This would be very similar to the current behavior of the Case Selector accepting errors with the two cases of "Error" and "No Error".)
The current behavior using "Not a Number/Path/Refnum" is very unintuitive. It requires the programmer to use Not Logic (i.e., do something if the reference is "not not valid").
I think the Array Element Gap should be sizable. This would facilitate lining up FP arrays with other items on the FP, or simply as a mechanism to add more apparent delineation between elements.
The size should be set in the Properties box, not by dragging the element gap with the mouse - that would add too much "cursor noise".
A new Property Node for this feature would complete Idea.