Fix syntax inside switch

[eelstork]

glog macros inside switch cause the MSVC compiler to emit error C2360.
I find that GCC may be lenient in this case; standards compliance should we require that we manually define scopes to declare and initialise variables within case statements.
Additionally, LOG(FATAL) doesn't exonerate us from the requirement to break after case.

[lukeyeager]

See #3130 as well.

[longjon]

Note that, by my quick read of https://github.com/google/glog/blob/master/src/glog/logging.h.in, it looks like this issue only affects LOG_FIRST_N.

This feels like an upstream issue to me; in general, macros should behave like function calls. We could consider a workaround (e.g., a wrapper around LOG_FIRST_N) pending an upstream fix (i.e., patch this in glog), but I'm not a big fan of mysteriously augmenting lots of calls that look perfectly fine.

Additionally, LOG(FATAL) doesn't exonerate us from the requirement to break after case.

Sure it does, just as return or exit(0) does.

[eelstork]

Raising an issue and suggesting a fix to the glog team.
In the meantime, I have updated the proposed fix to reflect suggested improvements.

[eelstork]

(closed by mistake; I wouldn't close this since I don't know whether it will be fixed in glog)

[eelstork]

With reference to google/glog#71 and google/glog#72, it appears that this can't be fixed in glog. @longjon, would you consider merging this?

[shelhamer]

@willyd is this patch still helpful for windows, and if so is #3130 even better for compatibility?

[eelstork]

Summary of what I know: patch is helpful; not fixable as an upstream issue (we tried); @longjon already debunked superfluous changes in #3130

[willyd]

@shelhamer I agree with @eelstork the patch is needed for windows compatibility. The current state of the windows branch looks more like #3130 but this PR has only the required changes so if it was my call I would merge this one and close #3130 without merging it. Current windows branch also as a comment explaining the new scope creation see this comment.

[shelhamer]

Thanks @eelstork for the fix, @longjon for the investigation, and @willyd for the current branch perspective.