To drive my point home, I shall now reveal what the compiler tells the errors
are. And boy, they are bogus.
type MyTuple = tuple(a: int, b: int)
# syntax...Error: invalid indentation
var myTuple: MyTuple = [a: 10, b: 10]
# syntax...Error: invalid indentation
type MyObject = ref Object
a: int
# syntax...Error: undeclared identifier: 'a'
var myObject: MyObject = MyObject(a = 5)
# syntax...Error: invalid expression: 'a = 5'
type MySpecialObject = ref object of MyObject
# syntax...Error: inheritance only works with non-final objects
var seqOfStrings: seq[int] = @[int]
# syntax...Error: internal error: expr(skType); unknown symbol
proc silentProc()* = discard
# syntax...Error: invalid indentation
proc inc(var a: int) = a += 1
# syntax...Error: ')' expected
The issue comes down to 3 problems for a nim novice:
* How can he trust the compiler (in production) when he cannot trust the
compiler error messages
* What's the use of top quality documentation when the compiler tells him to
search for the solution at wrong place
* How can he not be scared when a simple error triggers the compiler to
produce something like core dump (no example on this behaviour above)
Because there is no 1:1 correspondence between syntax errors and their semantic
consequences, this cannot be simply fixed by search/replacing existing error
messages. This also should/could not be fixed by complicating the parser (but
tinkering nimsuggest might be ok, I guess), because looking at the larger
context is not it's job. That's why I proposed a separate " _common syntax
error pattern detection tool_ ". I agree that this not something that the core
devs should be doing at the moment, but this would be a nice project for some
wannabe core dev out there. (And, eventually, a _quickfix on|off_ compiler
switch would be nice).
Fortunately detecting all random permutations is not necessary; I believe that
giving valid quick fix suggestions for the most common 20 syntax errors would
keep 80% of the beginners happy.
Let me propose a simple short term fix: When the error message is not to be
trusted, add a question mark to the error message. Instead of stately asserting
"Invalid intendation", the compiler should politely suggest: "Invalid
indentation?". This would save our beginner from spending his evening trying to
fix the error by prefixing/removing leading whitespace (been there, done that).
> But if you do find a crash please at least make sure you report it on GitHub.
Did that. Yesterday my first forum post, today my first issue, tomorrow the
world? When it comes to nim, I guess I'm now all in.
> the compiler's error message are constantly improving
True. To be honest, most of my beginner syntax errors I have collected long ago
produce valid error messages now as of 0.18. For example, one of my pain points
was confusing the usage of : and = operators. That the compiler now gives
relevant feedback on this confusion is to be highly appreciated by nim
beginners.
**Quiz 4** , strictly off-topic: try to guess the output
for x in [0..100]: echo x
