[various grammars] Improper use of start rule on RHS.

[kaby76]

EOF-terminated start rules are necessary for testing the grammars in this repo because they force the parser to read to the end of input and emit parse errors (#2405). In order to correct missing EOF-terminated start rules for grammars in this repo, various changes were made to correct the problem (e.g., #2675).

Generally, EOF-terminated start rules shouldn't be used on the right-hand side of any parser rule. Some of the rules labeled as start rules weren't implemented correctly, not taking into account whether the rule was used on the right-hand side of another parser rule. find-start.sh.txt

These grammars are suspected to have bad RHS use of a start rule.

• angelscript
• asn/asn_3gpp
• csharp (preprocessor)
• hypertalk
• javascript/ecmascript
• ./javascript/javascript
• ./javascript/jsx
• ./javascript/typescript
• ./kirikiri-tjs
• ./kotlin/kotlin-formal
• ./properties
• ./protobuf/protobuf3
• ./python/python
• ./python/python3_13
• ./sql/clickhouse
• ./sql/hive/v4
• ./sql/plsql
• ./sql/starrocks
• ./stringtemplate
• ./v

Some of these are probably ok, e.g., protobuf3, where I specifically call the parse twice (a symbol table is required for disambiguation of the grammar).

[kaby76]

There's something wierd about the EOF in closures that's not being checked consistently.

$ cat Eh.g4
grammar Eh;
start : eh+ EOF;
eh : 'eh?' | EOF;
WS : [ \t\n\r]+ -> skip;09/03-19:56:39 ~/eh/Generated-CSharp
$ make
bash build.sh
error(186): Eh.g4:2:0: rule start contains a closure with at least one alternative that can match EOF
mingw32-make: *** [makefile:3: build] Error 1
09/03-19:56:53 ~/eh/Generated-CSharp
$ pushd
~/eh2/Generated-CSharp ~/eh/Generated-CSharp
09/03-19:56:55 ~/eh2/Generated-CSharp
$ cat Eh2.g4
grammar Eh2;

start_: (statement semi)* EOF;
statement: 'eh?';
semi: NL+ | EOF;
NL: [\n\r];
WS: [ \t] -> skip;
09/03-19:57:01 ~/eh2/Generated-CSharp
$ make
bash build.sh
Restore complete (0.4s)

Build succeeded in 0.8s
Restore complete (0.4s)
  Test succeeded (0.7s) → bin\Debug\net8.0\Test.dll

Build succeeded in 1.7s

Workload updates are available. Run `dotnet workload list` for more information.
09/03-19:57:09 ~/eh2/Generated-CSharp
$

This check makes no sense: How is it possible to have a sentential form statement semi statement semi EOF -> statement EOF statement EOF EOF, where semi -> EOF? EOF is a really bad shorthand for { _input.LA(1) == -1 }?.