Bug: JRuby stacktraces with Java content can fail the reporting #47
Conversation
|
In my case, it was just a regular old LoadError which triggered this sad issue. |
olleolleolle
changed the title
|
Thanks! A PR would be greatly appreciated 👍 |
|
@mikker I have questions on desired behavior. I don't quite know how things behave with returned lists of Frame objects. If I add something like this" diff --git a/lib/opbeat/error_message/stacktrace.rb b/lib/opbeat/error_message/stacktrace.rb
index 6133f6f..0bb8704 100644
--- a/lib/opbeat/error_message/stacktrace.rb
+++ b/lib/opbeat/error_message/stacktrace.rb
@@ -27,9 +27,19 @@ module Opbeat
BACKTRACE_REGEX = /^(.+?):(\d+)(?::in `(.+?)')?$/.freeze
+ # regexp (optional leading X: on windows, or JRuby9000 class-prefix)
+ RUBY_INPUT_FORMAT = /
+ ^ \s* (?: [a-zA-Z]: | uri:classloader: )? ([^:]+ | <.*>):
+ (\d+)
+ (?: :in \s `([^']+)')?$
+ /x
+
+ # org.jruby.runtime.callsite.CachingCallSite.call(CachingCallSite.java:170)
+ JAVA_INPUT_FORMAT = /^(.+)\.([^\.]+)\(([^\:]+)\:(\d+)\)$/
+
class << self
def from_line config, line
- _, abs_path, lineno, function = line.match(BACKTRACE_REGEX).to_a
+ abs_path, lineno, function, _module_name = parse_line(line)
lineno = lineno.to_i
filename = strip_load_path(abs_path)
@@ -44,7 +54,21 @@ module Opbeat
private
+ def parse_line(unparsed_line)
+ ruby_match = unparsed_line.match(RUBY_INPUT_FORMAT)
+ if ruby_match
+ _, file, number, method = ruby_match.to_a
+ file.sub!(/\.class$/, '.rb')
+ module_name = nil
+ else
+ java_match = unparsed_line.match(JAVA_INPUT_FORMAT)
+ _, module_name, method, file, number = java_match.to_a
+ end
+ [file, number, method, module_name]
+ end
+
def strip_load_path path
+ return '' unless path
prefix = $:
.map(&:to_s)
.select { |s| path.start_with?(s) }Then the errors get quenched at least, but the new |
|
@mikker The older Rubies run first. Isn't that a little irritating? Or is it best? |
|
@mikker I need to push a JRuby test into the Travis matrix. I'll try to get just a single one in there. (And, I added it last, too. Ugh.) |
|
Apparently Sidekiq 5 can't do Ruby < 2.2. |
|
@mikker: (Hi! Yes, this got me going on this path. 2.1 is EOL'd. Can we stop supporting < 2.2?) |
|
We probably could. Can we get away with just freezing Sidekiq to be <5 in ruby 2.2 Gemfiles? |
|
@mikker Sure can start there. |
gemfiles/Gemfile.base
Outdated
| gem 'sidekiq', require: false | ||
| # Freeze Sidekiq to < 5 in Ruby 2.2 Gemfiles | ||
| if RUBY_VERSION >= '2' | ||
| gem 'sidekiq', (RUBY_VERSION >= '2.2.2' ? '~> 5' : '~> 4'), require: false |
There was a problem hiding this comment.
A beautiful line of code 😆 👍🏼
There was a problem hiding this comment.
Reminds us of the game "inte nudda golvet", "gulv brænder" etc.
Avoiiiiid the warnings.
|
@mikker In any event, whether or not this JRuby stacktrace feature can be implemented, the build improvements in this PR could be extracted to their own PR. |
JRuby stacktraces can contain lines that do not conform to the
BACKTRACE_REGEXpattern, which then returnsnilinstead of matches. Selection on these bad matches will fail with aNoMethodErrorfor#start_with?on nil.Gemfile.baseand tests that rely on itstrip_load_pathgets anilit returns""(Controversial? I don't know enough to know.)(Update): This is an Issue which was upgraded to a PR after the fact.
Update 2: This is not finished.
Current behavior
See error_message/stacktrace.rb for the code that fails.
Here's the current behavior: Opbeat throws a stacktrace and a "NoMethodError on nil".
Background
Update: The script below as a runnable file: https://gist.github.com/olleolleolle/2a9ba37a0fe774bd484929c0e0f155bb
Here is a way to make a Java exception in JRuby:
Having cut and pasted the
strip_load_pathandBACKTRACE_REGEXinto my irb, I could reproduce the issue with a standard Java error:Example of the stacktrace generated above:
Solution inspiration