__FL__ seems a bit obscure to me, should we maybe call this __FILELINE__? But I don't have a strong opinion, would be good if others chime in.

You have to reference it in every test so long is kind of cumbersome. For example:
let test_system n =
let r = FStar.Unix.system "echo HI" in
match r with
| WEXITED e -> if_test FL (e = 0) n None
| _ -> final_fail FL n None

I agree with Guido, I aslo find FL a bit obscure.
IMO, aggregating LINE and SOURCE_FILE in this formatting is quite opinionated and should probably be user-defined.

@briangmilnes why not using a tactic for that here, e.g.:

module Hello
open FStar.Char
open FStar.Tactics

let fl
  (x: unit)
  (#[let sealed_range = range_of_term (quote x) in
     let range = unseal sealed_range in
     let result = FStar.Range.explode range in
     exact (quote result)] loc: string * int * int * int * int)
 = let file, line_start, col_start, _line_end, _col_end = loc in
   file ^ "(" ^ string_of_int line_start ^ ")"

// This is line 14
let file_line: string = fl ()
let _ = assert_norm (file_line == "<input>(15)")

This is for logging and test packages. Nothing should be left to the user. In testing the ocaml wrap of unix and a few other things, I need the file/line 139 times. Terse is good here, although I eschew it in almost all situations.

A tactic that allows a function is a great alternative. The language server in emacs takes that and works but I can't get a command line to compile it. What's the change needed for compilation? It also has the advantage that the strings are not coming back from the lexer with strange UTF8.

Well, verbosity or terseness is a choice, right? I tend to prefer verbosity in such cases, but that's only my preference.

Nice! Here it won't work via the command line because of <input>: that's the file name given by the F* interactive protocol used by emacs I believe.
Running that from a file whose path is /tmp/Hello.fst, you should be able to assert_norm (file_line == "/tmp/Hello.fst(15)") (instead of the one I've put above).

The issue I have is it will not compile, not the string back from fl() or FL.

F*orge: VALIDATE _build/fstar/fst/checked/Hello.fst.checked

• Error 168 at src/Hello.fst(8,8-8,8):
  • Syntax error

1 error was reported (see above)
That's the (# line on my test of it running in OCaml.

The new single test here is not related to time at all? I think you can just drop Text.LexemeFL.fst in tests/micro-benchmarks and not add any new Makefiles.

validation-time is the name of the phase right? Nothing to do with time. I didn't put it in micro-benchmarks as it is a correctness test. For Unix I added a run-time test directory that I'm working on right now.

micro-benchmarks is a bad name really.. we use that directory for a whole bunch of unit tests. I'll rename it soon.

Using assert_norm breaks the dependency on tactics (and is maybe more idiomatic).

Also what is this testing? Seems to be just that 1) fl is nonempty and 2) that string_of_list and list_of_string are inverses.
(Note the missing 'e' from 'Test.LexemFL.fst(11)'.)

No what the darn string_of_list (list_of_string is doing is getting rid of oddball characters that prevent line 13 from working to show that FL is working. See the comment line above.

I have a commit pushed with the assert_norm changes. Do I need another PR?

So I have Test.FLLexeme moved to the directory to be renamed later, micro-benchmarks and validaton-time deleted and out of the makefile.

And this FStar revision will compile Luke's fl(), which is two characters shorter and more obvious than FL. I have no idea that reflection would do this for me as it is not in the book.

It would obviate this entire PR.

Guido which way do you want it? Lexeme or Tactics?