You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In my code, I'm implementing fmt.Formatter.
In order to write a string to the fmt.State, I'm calling io.WriteString().
I know that this function has an optimization that allows to call WriteString() if w implements it.
Sadly, in my case the "real" type of fmt.State is fmt.pp. fmt.pp only has Write().
in my case, it greatly improve performance and reduces memory allocation.
We could also implement WriteByte() and WriteRune(), since fmt.buffer has these methods.
io.WriteString() uses a runtime check so it looks like it should work without changing the State Interface definition. For go2 we could then think about wether it makes sense to update the State Interface.
@nikhita go ahead if you want too. Please add r@golang.org and moehrmann@google.com as reviewers. We can review before go1.10 but we can only submit when the code tree opens for go1.10 start of August.
@martisch
sorry, was busy at work - didn't have time to work on it.
But will try working on it on my flight to GopherCon now. :)
Maybe I can even ask any questions I have during the Go contributor workshop at GopherCon? :)
Following up here instead of on the CL discussion.
I would also favor to leave the internal buffer methods unexposed and only implement WriteString on pp as the proposal suggested. As far as i know there has been not yet been data or a request to add more io methods beyond WriteString. Keeping the api footprint low imposes less constraints on internals for future changes and avoids needing to implement them explicitly on pp later if buffer is replaced by something with different interfaces for internal optimization.
If it is decided to expose all the io methods from the buffer it would be good to make sure that there are no regressions due to the compiler not optimizing all the overhead away for the internal use of those methods (e.g. due to inline budget changes).
Activity
bradfitz commentedon Jun 24, 2017
Sounds reasonable.
nikhita commentedon Jun 25, 2017
I want to contribute to Go and noticed the
Help Wantedlabel. If no one is working on it, I'd like to take it! :)martisch commentedon Jun 25, 2017
io.WriteString() uses a runtime check so it looks like it should work without changing the State Interface definition. For go2 we could then think about wether it makes sense to update the State Interface.
@nikhita go ahead if you want too. Please add r@golang.org and moehrmann@google.com as reviewers. We can review before go1.10 but we can only submit when the code tree opens for go1.10 start of August.
nikhita commentedon Jun 25, 2017
@martisch will do that, thank you.
SCKelemen commentedon Jul 7, 2017
@SCKelemen
martisch commentedon Jul 10, 2017
@nikhita
did you have time to start your first CL for go?
Any blockers encountered that we can help with?
nikhita commentedon Jul 10, 2017
@martisch
sorry, was busy at work - didn't have time to work on it.
But will try working on it on my flight to GopherCon now. :)
Maybe I can even ask any questions I have during the Go contributor workshop at GopherCon? :)
martisch commentedon Jul 10, 2017
no problem. Asking at the contributor workshop sounds good.
gopherbot commentedon Jul 15, 2017
CL https://golang.org/cl/48871 mentions this issue.
rajathagasthya commentedon Sep 1, 2017
Just checking if this is still active. I see in the CL that it needs tests. I can help with that, if need be.
gopherbot commentedon Sep 4, 2017
Change https://golang.org/cl/61430 mentions this issue:
fmt: Implement pp.WriteString() methodstemar94 commentedon Sep 4, 2017
I mailed another way on how to resolve this. gopherbot will notice soon.
gopherbot commentedon Sep 4, 2017
Change https://golang.org/cl/61450 mentions this issue:
fmt: make buffer satisfy io interfacesgopherbot commentedon Sep 4, 2017
Change https://golang.org/cl/61451 mentions this issue:
fmt: make pp a composition with bufferrobpike commentedon Sep 7, 2017
Adding WriteString makes some sense, but please let's not open up an internal type to the full pressure of external I/O interfaces.
martisch commentedon Sep 7, 2017
Following up here instead of on the CL discussion.
I would also favor to leave the internal buffer methods unexposed and only implement WriteString on pp as the proposal suggested. As far as i know there has been not yet been data or a request to add more io methods beyond WriteString. Keeping the api footprint low imposes less constraints on internals for future changes and avoids needing to implement them explicitly on pp later if buffer is replaced by something with different interfaces for internal optimization.
If it is decided to expose all the io methods from the buffer it would be good to make sure that there are no regressions due to the compiler not optimizing all the overhead away for the internal use of those methods (e.g. due to inline budget changes).