Setting NAME field in DB file may break IOC

I can't think of any use case were modifying a SPC_NOMOD field would make sense. I think disallowing writes to these fields at database-load time would make sense.

Side-note: The Record Reference Manual says DTYP can NOT be modified (https://wiki-ext.aps.anl.gov/epics/index.php/RRM_3-14_dbCommon#Field_Summary_5) while dbCommon.dbd lists DTYP without "special(SPC_NOMOD)"? Am I missing something or is the Wiki wrong here?

Yup, changing NAME is a good way to break things. This changes dbCommon::name but not dbRecordNode::recordname nor the entry in the hash table of record names.

> I can't think of any use case were modifying a SPC_NOMOD field would make sense.

NELM and FTVL come immediately to mind.

> DTYP can NOT be modified ... Am I missing something or is the Wiki wrong here?

This inconsistency likely dates from the addition of the the somewhat esoteric extended device support feature, which allows changing DTYP. see 'struct dsxt' in devSup.h. mrfioc2 makes use of this.

Unless this is an alias dbRecordNode::recordname should be a pointer to dbCommon::name (we only free that storage when DBRN_FLAGS_ISALIAS is set). The reason changing the NAME field causes subsequent PV lookups to fail is that the record is now in the wrong gpHash bucket.

We do have to allow writes to SPC_NOMOD fields when loading the database, as Michael pointed out fields like NEML and FTVL are design fields that are SPC_NOMOD and we have to be able to set them during DB file parsing. The Wiki is out of date, fixes welcome (I can create an account for anyone who doesn't have one).

I would disallow writing to NAME in dbRecordField() in dbLexRoutines.c but strcmp() isn't the quickest way to identify the situation; NAME always has a field index of zero, so here's my fix to 3.14 which should merge up cleanly:

diff --git a/src/dbStatic/dbLexRoutines.c b/src/dbStatic/dbLexRoutines.c
index db95cc6..f9c2f8e 100644
--- a/src/dbStatic/dbLexRoutines.c
+++ b/src/dbStatic/dbLexRoutines.c
@@ -974,6 +974,12 @@ static void dbRecordField(char *name,char *value)
        yyerror(NULL);
        return;
     }
+ if (pdbentry->indfield == 0) {
+ epicsPrintf("Can't set \"NAME\" field of record \"%s\"\n",
+ dbGetRecordName(pdbentry));
+ yyerror(NULL);
+ return;
+ }
     dbTranslateEscape(value, value); /* yuck: in-place, but safe */
     status = dbPutString(pdbentry,value);
     if(status) {

Looks good to me.
(On my phone, so by visual inspection only.)