Experimental Physics and
Industrial Control System

At ESS we have the convention to have  in a particular position of the record name,
the '#' character for "internal" record, i.e. a record  which is used for instance for intermediate calculations,
In practice this internal record should not be used outside the IOC,
i.e. no channel finder, no archiving , no alarm, no autosave, and so on.
We use as well the '-' and ':'  characters to separate some record name part, e.g.

"Normal" Record Name

Sys-Sub:Dis-Dev-Idx:Prop

"Internal" Record Name

Sys-Sub:Dis-Dev-Idx:#Prop

'#', ':', '-' as well as '_' are 'legal' or 'illegal' characters ?

Can we use them without the possible risk to have EPICS crashing in any future release ?

-----Original Message-----
From: Tech-talk <tech-talk-bounces at aps.anl.gov> On Behalf Of Michael Davidsaver via Tech-talk
Sent: Saturday, June 13, 2020 5:08 PM
To: Hu, Yong <yhu at bnl.gov>
Cc: EPICS Tech-Talk <tech-talk at aps.anl.gov>
Subject: Re: weird record names?

On 6/10/20 9:13 AM, Michael Davidsaver wrote:

On 6/9/20 6:33 PM, Hu, Yong wrote:

Michael,

When you say "restrictions", does that mean a newer version of EPICS IOC will fail to load a .db file in which a record name contains a "weird" character?

Eventually, yes.  That is what I'm considering.

cf. for a concrete proposal.  https://github.com/epics-base/epics-base/pull/78

If the IOC fails to load .db because it does not follow the new rules/restrictions, facilities that have already used "weird" characters will suffer badly when they upgrade EPICS base to a newer version.

I recognize and want to avoid this.  
Any new restriction would begin as a warning, and likely remain so for
a number of years.
This would mainly serve as an early indication of names which would
cause problems in other contexts.  Record names which would be
impossible in link to (like "a b" below, or "123"), or which would
never parse as channel names (like "x.y").

3.15.0.1 introduced a warning for record names containing " \"'.$".  
These might now begin to be enforced.  And include "\t" for good
measure.

Yong


On 6/8/20, 5:09 PM, "Tech-talk on behalf of Michael Davidsaver via Tech-talk" <tech-talk-bounces at aps.anl.gov on behalf of tech-talk at aps.anl.gov> wrote:

   I'm looking to collect examples of epics record names in the wild.

   This is an early step towards (maybe) adding restrictions on what
   characters a record name can contain, and in what positions.
   eg. restricting possible first and last characters.

   In particular I'm looking for examples including characters
   beyond the usual alphanumeric separated by ':' or '-'.
   And in what positions they may (or may not) appear.

   An example from the NSLS2 naming convention (for which I am have
   some responsibility).

TST{evm:1D-DlyGen:31}Evt:Trig2-SP

   Which is notable for containing "{" and "}" in the middle.



   As background.  At present, records can be created with almost
   any characters in a name.

record(ai, "a b") {}
record(ai, "x.y") {}
record(ai, "x\"") {}

   However, record names including '.' can't be addressed and aren't
   very useful.  Further, names including spaces can't be targeted
   by links.

   On the subject of links.  Dirk was surprised to find that
   the syntax for link parsing treats "[0]" as a record name,
   but "[0,1]" as a constant array.


https://urldefense.com/v3/__https://bugs.launchpad.net/epics-base/*bu
g/1882520__;Kw!!P4SdNyxKAPE!TwNJ4RpksYZmN8YbfgjiN7Y3igc4486VOLcj-_1L1
KUqPcMdbvxVXH_bSrI3$

   Rather than carving out another exception, I'd like to look at
   coding a general rule.