Serial Modbus master reopen may always fail after error #71
Comments
rtollert
added a commit
to rtollert/ModbusModules
that referenced
this issue
Aug 21, 2019
Fix for LabVIEW-DCAF#71. modbus initialize API object.vi: - New design assertions: 1. Refcounts (Clients Count DVRs) are never closed. 2. All API objects participate in refcounting, not just serial masters. This does not affect the behavior of things that aren't serial masters, and dramatically simplifies logic both here and in close.vi. - Accordingly, the FGV variant now holds (Modbus API, refcount) tuples instead of (Modbus Master, refcount) tuples. And it is notionally getting indexed by (class, address) tuples, e.g. ("serial master", VISA instrument), or ("TCP master", IP address, port). close.vi: - Unwire "error in" from Decrement Clients Count.vi and the rest of that shutdown chain, to ensure that it always runs. This helps preserve normal close semantics for this VI. - Remove Delete DVR call to Clients Count DVR, because modbus initialize API object.vi is now guaranteed to hold it open. Add a brief comment to this effect. Reorder final Merge Errors call accordingly. Build lookup key for configuration.vi: - Builds the strings of (class, address) tupes which are used to index into the FGV variant in modbus initialize API object.vi. - "ok?" indicator controls whether or not reusing the modbus instance from the FGV is permitted -- false for everything but serial masters. (Refcount DVRs are always reused.)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
#63 (the fix for #59) introduced a DVR ("Client Count DVR") for refcounting serial master instances, to defer closing the underlying VISA session. This was implemented with a variant FGV mapping VISA instrument name to (instance, DVR) tuple in
Modbus Master.lvclass:modbus initialize API object.vi. Onclose.vi, the DVR is deleted once it goes to zero. However, the FGV is never informed that the DVR has been deleted.Now, normally this is fine, because of the "Not a refnum?" check: if all masters really did close, then this will return FALSE for the VISA session, thus forcing the creation of a new DVR anyway. However, the decrement-refcount-and-close process in
close.vialso wires in error in. Suppose that there is a real error wired into error in ofclose.vi, with one serial master open. Then the subsequent sequence of events will proceed as follows, AFAIK:Decrement Clients Count.viwill skip because it implements standard error out functionality. The refcount returned will always be zero (even if the port has been opened reentrantly).Modbus Master.lvclass:Shutdown.viwill effectively skip because of the underlying code. (Careful readers will need to trust me on this; some of the underlying code is password-protected.)modbus initialize API object.vi. The variant lookup will succeed, and because the VISA session is still open, the VI will continue down the TRUE case of "serial master", where the instance and refcount DVR are reused.Increment Clients Count.viis attempted. This will fail, because the DVR no longer exists.close.vi. Lather, rinse, and repeat.Note, the precise details of this issue have only been inferred by inspection; I have not yet attempted a minimal reproducing case. I have sketched out an untested fix.
@agomezni @becega
The text was updated successfully, but these errors were encountered: