I can use times.epochTime() in func.

[demotomohiro]

I write a func that call times.epochTime() and Nim compiles it without error or warning.

Example

import times

func testEpo(x: float): float = epochTime() + x

echo testEpo(1.0)

Current Output

1580491911.732826

Expected Output

Error: 'testEpo' can have side effects

According to Araq in IRC, "noSideEffect = proc result only depends on its parameters = proc doesn't use global state".

Using times.cpuTime in func show compile error.

import times

func testCPU(x: float): float = cpuTime() + x

echo testCPU(1.0)

Output:

test.nim(3, 6) Error: 'testCPU' can have side effects

>nim -v
Nim Compiler Version 1.1.1 [Windows: amd64]
Compiled at 2020-01-31
Copyright (c) 2006-2019 by Andreas Rumpf

git hash: adc52b003960817553aa3481e21a9b335a504e7f
active boot switches: -d:release

[timotheecour]

you can try to submit a fix by annotating the root cause of the effect (the importc proc), then it should bubble up to callers

[demotomohiro]

times.epochTime() on windows calls getSystemTimeAsFileTime.
Adding sideEffect pragma to getSystemTimeAsFileTime fix this issue.

But I have some questions.
times.epochTime() has tags: [TimeEffect] pragma but Nim doesn't count it as the effect?
Because Nim compiler doesn't know whether importc procs have sideEffect, Nim should assume importc procs have a sideEffect unless it has noSideEffect pragma, isn't it?
(But it can be breaking change.)

[Araq]

Because Nim compiler doesn't know whether importc procs have sideEffect, Nim should assume importc procs have a sideEffect unless it has noSideEffect pragma, isn't it?

No, the default should be "assume .noSideEffect" for C functions, most of them are in libraries and need to be re-entrant so assuming that works better out, all things considered. IMO.

[demotomohiro]

If people forgot to add sideEffect pragma to an importc proc that actually have a side effect and wrote a func that call it, compiler would show no errors. In this case, people might use that func for long time without realizing that the func actually has a side effect.
If the default of importc were assumed sideEffect and forgot to add noSideEffect pragma to importc proc, compiler show error when the importc proc is used in a func. So I can check all procs called from the func and add a noSideEffect pragma to the importc proc.

[Araq]

Good point but who is gonna update millions of lines of wrapper code?