> After some iterations, we merged the branch with the new build scheme.

the commit to implement this change in the manalyzer was not pushed, for reasons unknown.

fixed, commit f2b4dc87ca4830f6bed8667d6a4ee4afd6d242a1

K.O.

> I have a manalyzer that uses a derived class of TMFeRpcHandlerInterface to communicate information to 
> Midas during online running.  At the end of each run it saves out custom data in the 
> TMFeRpcHandlerInterface::HandleEndRun override. This works really well.
> However, when I run offline on a Midas output file the HandleEndRun method is never called and my data is 
> never saved.  Is this intentional?  I understand that there is no point for the HandleBinaryRpc method offline, 
> but the other methods (HandleEndRun, HandleBeginRun etc) could serve a purpose.  Or is it a conscious 
> choice to ignore all of TMFeRpcHandlerInterface when offline?

apologies for delayed response.

I saw the question, completely did not understand it, only now got around to figure out what is going on.

according to manalyzer/README.md, section "manalyzer module and object life time", BeginRun() and EndRun() is called 
always. Offline and online. What you see would be a bug that we do not see in our environment. I confirm this by 
running manalyzer in a demo mode: ./bin/manalyzer_test.exe  --demo -e10 -t

no, wait, you say you use HandleBeginRun() and HandleEndRun(). this is not right, they are not part of the manalyzer 
API and they indeed are only used when running online.

correct solution would be to use BeginRun() and EndRun() instead of HandleBeginRun() and HandleEndRun().

you could also save your data in the module destructor (although good programming recommendation is to use 
destructor only for unavoidable things, like freeing memory, etc).

K.O.

I think I added the cache size correctly:

  {"Trigger",               /* equipment name */
      {1, 0,                 /* event ID, trigger mask */
         "SYSTEM",           /* event buffer */
         EQ_POLLED,          /* equipment type */
         0,                  /* event source */
         "MIDAS",            /* format */
         TRUE,               /* enabled */
         RO_RUNNING |        /* read only when running */
         RO_ODB,             /* and update ODB */
         100,                /* poll for 100ms */
         0,                  /* stop run after this event limit */
         0,                  /* number of sub events */
         0,                  /* don't log history */
         "", "", "", "", "", // frontend_host, name, file_name, status, status_color
         0, // hidden
         0  // write_cache_size <<--------------------- set this to zero -----------
      },
   }

K.O.

the main purpose of the event buffer write cache is to prevent high contention for the 
event buffer shared memory semaphore in the pathological case of very high rate of very 
small events.

there is a computation for this, I have posted it here several times, please search for 
it.

in the nutshell, you want the semaphore locking rate to be around 10/sec, 100/sec 
maximum. coupled with smallest event size and maximum practical rate (1 MHz), this 
yields the cache size.

for slow control events generated at 1 Hz, the write cache is not needed, 
write_cache_size value 0 is the correct setting.

for "typical" physics events generated at 1 kHz, write cache size should be set to fit 
10 events (100 Hz semaphore locking rate) to 100 events (10 Hz semaphore locking rate).

unfortunately, one cannot have two cache sizes for an event buffer, so typical frontends 
that generate physics data at 1 kHz and scalers and counters at 1 Hz must have a non-
zero write cache size (or semaphore locking rate will be too high).

the other consideration, we do not want data to sit in the cache "too long", so the 
cache is flushed every 1 second or so.

all this cache stuff could be completely removed, deleted. result would be MIDAS that 
works ok for small data sizes and rates, but completely falls down at 10 Gige speeds and 
rates.

P.S. why is high semaphore locking rate bad? it turns out that UNIX and Linux semaphores 
are not "fair", they do not give equal share to all users, and (for example) an event 
buffer writer can "capture" the semaphore so the buffer reader (mlogger) will never get 
it, a pathologic situation (to help with this, there is also a "read cache"). Read this 
discussion: https://stackoverflow.com/questions/17825508/fairness-setting-in-semaphore-
class

K.O.

> the main purpose of the event buffer write cache

how to control the write cache size:

1) in a frontend, all equipments should ask for the same write cache size, both mfe.c and 
tmfe frontends will complain about mismatch

2) tmfe c++ frontend, per tmfe.md, set fEqConfWriteCacheSize in the equipment constructor, in 
EqPreInitHandler() or EqInitHandler(), or set it in ODB. default value is 10 Mbytes or value 
of MIN_WRITE_CACHE_SIZE define. periodic cache flush period is 0.5 sec in 
fFeFlushWriteCachePeriodSec.

3) mfe.cxx frontend, set it in the equipment definition (number after "hidden"), set it in 
ODB, or change equipment[i].write_cache_size. Value 0 sets the cache size to 
MIN_WRITE_CACHE_SIZE, 10 Mbytes.

4) in bm_set_cache_size(), acceptable values are 0 (disable the cache), MIN_WRITE_CACHE_SIZE 
(10 Mbytes) or anything bigger. Attempt to set the cache smaller than 10 Mbytes will set it 
to 10 Mbytes and print an error message.

All this is kind of reasonable, as only two settings of write cache size are useful: 0 to 
disable it, and 10 Mbytes to limit semaphore locking rate to reasonable value for all event 
rate and size values practical on current computers.

In mfe.cxx it looks to be impossible to set the write cache size to 0 (disable it), but 
actually all you need is call "bm_set_cache_size(equipment[0].buffer_handle, 0, 0);" in 
frontend_init() (or is it in begin_of_run()?).

K.O.

> > the main purpose of the event buffer write cache
> how to control the write cache size:

OP provided insufficient information to say what went wrong for them, but do try this:

1) in ODB, for all equipments, set write_cache_size to 0
2) in the frontend equipment table, set write_cache_size to 0

That is how it is done in the example frontend: examples/experiment/frontend.cxx

If this configuration still produces an error, we may have a bug somewhere, so please let us know how it shakes out.

K.O.

> time to choose the next c++!

Ununtu-24.04, MIDAS builds with -std=c++23 without errors or any additional warnings. (but does it work? "make 
test" is ok).

K.O.

Andrea Capra reported a problem with manalyzer module initialization order.

Original manalyzer design relies on the fact than module static constructors run in the same order as they 
appear on the linker command line. This has been true for a very long time, but now we have evidence that on 
Rocky Linux (unknown gcc version), this is no longer true.

The c++ standard does not define any specific order for static constructor in different source files. 
(unlike C, C++ tends to treat the linker as something magical and ask the linker to do magical things).

The tmfe c++ modular frontend was designed after manalyzer and one design change is to explicitly construct 
all objects from main(). (at the acceptable cost of extra boiler plate code).

One solution is to use GCC attribute "init_priority (priority)", see
https://stackoverflow.com/questions/211237/static-variables-initialisation-order
https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Attributes.html

To use this, manalyzer module registration should be modified to read:

static TARegister tar __attribute__((init_priority(200))) (new ExampleCxxFactory);

A less magical solution is to change manalyzer design similar to tmfe where modules are constructed and 
registered explicitely in the order specified by source code.

K.O.

P.S. I now ran out of time to test this and commit it to the documentation. It also need to be tested with 
LLVM C++ compilers.

> > Hello,
> > 
> > Does anyone have experience with writing a MIDAS frontends to communicate with a device that operates using LabView (e.g. superconducting magnets, cryostats etc.). Any information or experience regarding this would be highly appreciated.
> > 
> > thanks,
> > Zaher
> 
> We do have a superconducting magnet from Cryogenic, UK, which comes with a LabView control program on a Windows PC. I did the only reasonable with this: trash it in the waste basket. Do NOT use Labveiw for anything which should run more than 24h in a row. Too many bad experiences with LabView control programs 
> for separators at PSI and other devices. Instead of the Windows PC, we use MSCB devices and RasperryPis to communicate with the power supply directly, which has been proven to be much more stable (running for years without crashes). I'm happy to share our code with you.
>

Our parallel experience with the CERN ALPHA anti-hydrogen experiment: they have developed a whole labview empire
to control the cryogenics, the magnets, the positron source, the anti-proton trap, anti-hydrogen trap, etc.

At some point there was a wall of monitors in the counting room - each labview computer controlled one or two things -
so there is very many computers and each had to have a monitor (and mouse and keyboard).

All the data from this labview empire is logged to MIDAS history via felabview and feGEM, and they use
the MIDAS history to look and monitor almost everything. Control is done via Labview and Labview
based FPGA sequencers (National Instruments PXI hardware, $$$$$).

This works reasonably well to publish several papers in Nature.

But not 100%:

1) difficulties with labview source control (cannot be trivially managed by git, I guess)
2) unending fight against Microsoft and CERN IT trying to reboot the computers at the wrong time
3) more recently, forced Microsoft updates require trashing perfectly good machines and buy new ones

At TRIUMF there is very little Labview. All experiments use MIDAS and EPICS for most things.

Based on this experience, I agree with Stefan, today's sweet spot is RaspberryPi machines with USB attached
gizmos to control stuff. On the software side, drive the mess with MIDAS and custom web pages.

K.O.

> > bitbucket automatic builds
> 
> Unfortunately we will break the automatic build each time a program outputs one different character, which even might happen if we add a line of code and 
> a cm_msg() gets produced with a different line number. Is there a standard way to update testexpt.example (like "make testexpt" or so). Should be trigger 
> the update of testexpt.example before each commit via a hook? 
> 

Actually line numbers are not logged by messages printed from "make test", so moving code around does not break the test.

Changing what programs output does break the test and this is intentional - somebody must look at confirm
that program output was changed on purpose or because a bug was introduced (or fixed).

Most "make test" things work this way - run programs, compare output to what is expected. Discrepancies are flagged for human examination.

K.O.

> > > currently to link Midas to project one has to do several steps ...
> > this information is incorrect. please read https://daq00.triumf.ca/elog-midas/Midas/2258
> 
> I admit that i did not see your post about targets import
> via `include($ENV{MIDASSYS}/lib/midas-targets.cmake)`
> before implementing changes to cmake scripts.
>
> But in this respect the way you propose to do it via `include` should still work.
>

I proposed nothing, you did the proposing. I spent many hours trying to understand cmake (mission 
impossible!) and many more hours to implement the previously existing package scheme based
on the cmake "EXPORT" function.

> Note however that `include(...)` way is very unusual as one have to know exactly
> where `...-targets.cmake` is located and standard way in cmake is via `find_package`
> (similar to how e.g. ROOT, Geant4, etc. are found and linked).

Very difficult to cut-and-paste "include($ENV{MIDASSYS}/lib/midas-targets.cmake)".

You cannot simplify-out $ENV{MIDASSYS} because computer cannot read your mind, which of the 10 copies
of midas you want to use from which user account on which day.

Argument about "very unusual" I would buy, I am not a cmake expert and I do not know which package
finding method is in favour today.

>
> Though i again admit that maybe the namespace change was a bit too much as it may
> have broken previous users of `include($ENV{MIDASSYS}/lib/midas-targets.cmake)`
> 

I believe it did break at least one experiment, after updating MIDAS to latest version,
the analyzer would not build.

Speaking of which, did you implement your new scheme for the manalyzer so that it works
in standalone mode (without MIDAS)?

If you did not, now we have two schemes, your new scheme just for MIDAS and my old scheme
for manalyzer *and* MIDAS. xkcd 927.

> 
> - shortcomings of what was before is usage of non-standard `include(...)`
>

You should have started by posting a message spelling it out: Konstantin implemented
a scheme that uses the cmake "export" function to find midas, mfe and manalyzer,
it is very nice and works ok, but it is non-standard/obsoleted/obscure/frowned-upon/
unpopular/I-do-not-like-it/I-did-not-invent-it, and I propose implementing a new scheme
based on find_package().

>
> - one shortcoming i see for new implementation is usage `midas::` namespace
>   (mentioned above) that may have broken some setups
> 

If you think that your changes will break other people code, you should explicitely
say this in a message to this forum and hopefully provide instruction on fixing it,
i.e. in your makefile, please replace "midas" with "midas::midas".

> 
> - `find_package` is standard and recommended way of finding packages
>

Do you have a reference for this? When I look at cmake documentation, I do not see
any specific recommendation on creating packages and finding them. I do see
other people's code for finding packages and often spend hours fighting
them because said methods are designed to work only on the developer's laptop.

P.S. Did anybody ask Ben to update the MidasWiki documentation with the new find_package() information?

K.O.

> > > ALL MIDAS CLIENTS GET RE-COMPILED after the new code is applied.
>
> - I did clearly announce this change in the forum.
>

I missed the announcement and I was surprised to discover this change.

This is why I changed the title from "useful improvement" to "ACHTUNG!!!"

> - The extension is not there "just for fun" but needed by several experiments which experience spurious alarms 
> triggered by some noisy signals.

Agreed, a useful improvement.

To prevent this kind of trouble in the future, I think I should finish my crusade against db_get_record().

> - If there are frontend programs which have not been re-compiled for ten years, I think it is very unlikely that these 
> experiments need the latest cutting edge version of midas. The can safely stick to your tag midas-2024-12-a and not 
> experience the issue of different /Alarm trees.

Yes, now that I added the tag, posted a more visible notice and people who have to run frontends build against older midas 
(because reasons) can breathe out.

Also please note that "Konstantin != all experiments at TRIUMF".

K.O.

> In the PSI muSR laboratory, we are running about 140 slow control devices across six instruments using Format FIXED.
> Could you please wait a bit with removing support for it so that we can assess if/how this will affect us?

Roger. I believe as long as these data do not go into the MIDAS event data stream, you should not see any difference from 
changing "FIXED" to "MIDAS", I hope you can test it and report how it works for you. If you see that things change in ODB 
or in history, perhaps we can implement some kind of workaround to make the transition transparent.

K.O.

> Andrea Capra reported a problem with manalyzer module initialization order.

Permanent solution is now implemented.

In the analyzer module constructor (TARunObject), please set the value of fModuleOrder (default value is 0).

Modules with smaller fModuleOrder will run first, modules with bigger fModuleOrder will run last.

Please use the value range between -1 (built-in EventDump module) and 9999 (built-in InteractiveModule).

Modules with equal fModuleOrder will run in the same order as they were registered (std::stable_sort).

Example manalyzer modules and documentation README.md have been updated.

Linker command line and GCC attribute methods should also work, but I thought it better to provide explicit 
programmatic control of module ordering.

P.S. the c++ tmfe frontend module design is newer and the problem of module (equipment) ordering is solved by 
constructing them explicitly in main().

manalyzer commit:

commit 6ca130808cd05ead734450391bf6defc8335c04a (HEAD -> master, origin/master, origin/HEAD)
Author: Konstantin Olchanski <olchansk@daq00.triumf.ca>
Date:   Tue Mar 25 15:53:13 2025 -0700

    implement TARunObject::fModuleOrder to specify required ordering of analyzer modules

K.O.

> The question was about the TMFeRpcHandlerInterface, not the TARunObject interface.  Derived classes of TARunObject do indeed work as expected in our 
> environment.  We have worked around the issue by using an implementation of TARunObject as well as the (separate) implementation of 
> TMFeRpcHandlerInterface.

then I do not understand the question. TMFeRpcHandlerInterface stuff is only used when running online and connected to MIDAS. How does it come into the 
picture when you analyze a data file offline? ProcessMidasOnlineTmfe() does not run, the RpcHandler object is not constructed.

maybe if you point me to your source code, I can see what you are doing?

K.O.

> https://cmake.org/cmake/help/latest/guide/using-dependencies/index.html#guide:Using%20Dependencies%20Guide

thank you for providing a link to latest cmake find_package() guide.

I notice that this documentation was added in cmake 3.24 released circa Nov 2022
and does not exist in older versions. (It is easy to see in git history the last
time I touched any cmake stuff in midas).

I still see no documentation on "this is how you write a package that other people can import
using find_package()". Only see documentation on how to use find_package() on packages
that somebody who somehow knows how to do it already wrote.

> In the original message to this thread i posted reference to PR
> (https://bitbucket.org/tmidas/midas/pull-requests/48)

this pull request was rail-roaded through during the holidays without any
discussion on this forum. I was not given an opportunity to comment to it,
it was pushed and merged faster than I could blink.

bottom line. I voted against using cmake and was over-ruled. To me this cmake stuff
is only a source of wasted time and created bad feelings.

if midas is required to use cmake, we should have somebody on the team that at least
understands it and if not love it, at least does not hate it.

K.O.

Confirming warning is fixed. Thanks! K.O.

> > Unnamed person who added this clever bit of c++ coding, please fix this compiler warning. Stock g++ on Ubuntu LTS 24.04. Thanks in advance!
> > 
> > /home/olchansk/git/midas/src/system.cxx: In function ‘std::string ss_execs(const char*)’:
> > /home/olchansk/git/midas/src/system.cxx:2256:43: warning: ignoring attributes on template argument ‘int (*)(FILE*)’ [-Wignored-attributes]
> >  2256 |    std::unique_ptr<FILE, decltype(&pclose)> pipe(popen(cmd, "r"), pclose);
> > 
> > K.O.
> 
> Replace the code with:
> 
>    auto pclose_deleter = [](FILE* f) { pclose(f); };
>    auto pipe = std::unique_ptr<FILE, decltype(pclose_deleter)>(
>       popen(cmd, "r"),
>       pclose_deleter
>    );
> 
> 
> Hope this is now warning-free.
> 
> Stefan

> I hopefully fixed the waring (narrowing down from size_t to int). Please double check with your compiler.

Nope, it turns out complain is about the read() size argument, they really really really want it to be 
size_t.

Looks like correct sequence is:

size_t size = file_size_somehow();
char* buf = (char*)malloc(size+1);
ssize_t rd = read(fd, buf, size);
if (rd < 0) { complain(); return; } // read() error
if ((size_t)rd != size) { complain(); return; } // cast to size_t is safe after we know rd >= 0
buf[rd] = 0; // make sure data is NUL terminated

What a mess. I want my UNIX V7 and K&R C back!

Perhaps I should add in system.cxx - ss_read_file(const char* filename, std::vector<char> &data);

Fixed it in mscb.cxx, odbedit.cxx, mhttpd.cxx and msequencer.cxx, commited, pushed.

U-24 builds without warnings. fwew...

K.O.

>  > All this is kind of reasonable, as only two settings of write cache size are useful: 0 to 
> > disable it, and 10 Mbytes to limit semaphore locking rate to reasonable value for all event 
> > rate and size values practical on current computers.
> 
> Indeed KO is correct that only 0 and 10MB make sense, and we cannot mix it. Having the cache setting in the equipment table is 
> cumbersome. If you have 10 slow control equipment (cache size zero), you need to add many zeros at the end of 10 equipment 
> definitions in the frontend. 
> 
> I would rather implement a function or variable similar to fEqConfWriteCacheSize in the tmfe framework also in the mfe.cxx 
> framework, then we need only to add one line llike
> 
> gEqConfWriteCacheSize = 0;
> 
> in the frontend.cxx file and this will be used for all equipments of that frontend. If nobody complains, I will do that in April when I'm 
> back from Japan.

Cache size is per-buffer. If different equipments write into different event buffers, should be possible to set different cache sizes.

Perhaps have:

set_write_cache_size("SYSTEM", 0);
set_write_cache_size("BUF1", bigsize);

with an internal std::map<std::string,size_t>; for write cache size for each named buffer

K.O.

I do not understand what you are doing. If you are offline, there is no TMFE singleton instance,
there is nothing TMFeRpcHandlerInterface to attach to, there is nobody to call TMFeRpcHandlerInterface methods.

Maybe what you are asking for is a mode where you analyze data from a file, but you want your analysis code
to think that it is online and that it is analyzing live data. This requires creating a fake TMFE singleton, attaching
TMFeRpcHandlerInterface to this fake TMFE singleton and using ProcessMidasOnlineTmfe(), driven by all this fake stuff:
a fake OpenBuffer() that actually opens a file, a fake ReceiveEvent() that actually reads from a file, fake callbacks
for begin and end run, etc.

That's a lot of work, but for what purpose? What is it about the existing offline and online modes that you do not like
and how all this fake stuff will make it better for you?

P.S. This is a 3rd version of my reply. Wrote and deleted 2 version. I think I completely do not understand
what you are doing and you completely do not understand what I am saying. Communication is not happening.

P.P.S. Simplest if you show me your code (email, elog), I am quite good at reading code and divining what
people are trying to do. You do not have to show me any of your secret secret stuff.

K.O.

> This was exactly the question, should I expect it to run?  There's no point in the HandleBinaryRpc method offline, but there's an argument that the HandleBeginRun/HandleEndRun methods have a use.
> I have the answer and we have a workaround, thanks.
> 
> > then I do not understand the question. TMFeRpcHandlerInterface stuff is only used when running online and connected to MIDAS. How does it come into the 
> > picture when you analyze a data file offline? ProcessMidasOnlineTmfe() does not run, the RpcHandler object is not constructed.
> > 
> > maybe if you point me to your source code, I can see what you are doing?
> > 
> > K.O.