Allow print() to be used for debugging analysis server.

Allow print() to be used for debugging analysis server.

We now run analysis server inside a zone so that all print() calls it
executes can be captured to a buffer (rather than being sent to
standard output, where they will confuse the client).  When the server
is run in HTTP server mode, the print buffer is displayed on the
status page.

R=brianwilkerson@google.com, danrubel@google.com

Committed: https://code.google.com/p/dart/source/detail?r=41380

[Paul Berry, 2014-10-28 20:51:34]

paulberry@google.com changed reviewers:
+ brianwilkerson@google.com

[scheglov, 2014-10-28 20:59:28]

DBC

I don't like this.
It mean that now we cannot just see output from the server in Eclipse, with mix
of debug output and corresponding requests/responses, and have to open browser
and refresh it.
Merging debug output and requests/responses is up to the developer then.

[danrubel, 2014-10-28 21:01:02]

danrubel@google.com changed reviewers:
+ danrubel@google.com

[danrubel, 2014-10-28 21:01:02]

LGTM. Do we want to pipe the print lines into debug notifications so that you
can see interaction in chronological order?

https://codereview.chromium.org/686763004/diff/1/pkg/analysis_server/lib/http...
File pkg/analysis_server/lib/http_server.dart (right):

https://codereview.chromium.org/686763004/diff/1/pkg/analysis_server/lib/http...
pkg/analysis_server/lib/http_server.dart:23: static const int
PRINT_BUFFER_LENGTH = 1000;
MAX_PRINT_BUFFER_MAX_LENGTH ?

[danrubel, 2014-10-28 21:02:37]

On 2014/10/28 20:59:28, scheglov wrote:
> DBC
> 
> I don't like this.
> It mean that now we cannot just see output from the server in Eclipse, with
mix
> of debug output and corresponding requests/responses, and have to open browser
> and refresh it.
> Merging debug output and requests/responses is up to the developer then.

I agree with you point. Perhaps pipe print calls into a new debug notification?

[Paul Berry, 2014-10-28 21:06:58]

On 2014/10/28 20:59:28, scheglov wrote:
> DBC
> 
> I don't like this.
> It mean that now we cannot just see output from the server in Eclipse, with
mix
> of debug output and corresponding requests/responses, and have to open browser
> and refresh it.
> Merging debug output and requests/responses is up to the developer then.

The problem with what we currently have is that if there's a print() statement
in the analysis server code, it causes the analysis server to violate the
protocol spec (because it outputs text that isn't proper JSON).  Currently the
editor handles that ok, by detecting when text from the server isn't valid JSON
and ignoring it, but I don't think we should rely on clients to do that.

How do you feel about Dan's proposal to pipe print calls into a new debug
notification?  I could easily do that as a follow-up to this CL.

[Brian Wilkerson, 2014-10-28 21:17:38]

LGTM

> How do you feel about Dan's proposal to pipe print calls into a new
> debug notification?

I think we've discussed this before, and I like the idea. I'd like to have us
define a debug domain and for Editor to subscribe for debugging output
notifications. But I also like the idea of having a buffer that lets us show the
output in the browser.

[scheglov, 2014-10-28 22:19:11]

On 2014/10/28 21:17:38, Brian Wilkerson wrote:
> LGTM
> 
> > How do you feel about Dan's proposal to pipe print calls into a new
> > debug notification?
> 
> I think we've discussed this before, and I like the idea. I'd like to have us
> define a debug domain and for Editor to subscribe for debugging output
> notifications. But I also like the idea of having a buffer that lets us show
the
> output in the browser.

lgtm

[Paul Berry, 2014-10-28 23:29:14]

https://codereview.chromium.org/686763004/diff/1/pkg/analysis_server/lib/http...
File pkg/analysis_server/lib/http_server.dart (right):

https://codereview.chromium.org/686763004/diff/1/pkg/analysis_server/lib/http...
pkg/analysis_server/lib/http_server.dart:23: static const int
PRINT_BUFFER_LENGTH = 1000;
On 2014/10/28 21:01:02, danrubel wrote:
> MAX_PRINT_BUFFER_MAX_LENGTH ?

Done.

[Paul Berry, 2014-10-28 23:29:33]

Committed patchset #1 (id:1) manually as 41380 (presubmit successful).