We should get out of the business of parsing backtraces and only use
backtrace locations. Backtrace locations have the file and line number
information baked in, so we don't need to parse things anymore
This commit adds a SyntaxErrorProxy object to active support and wraps
syntax error exceptions with that proxy object. We want to enhance
syntax errors with information about the source location where they
actually happened (normally the backtrace doesn't contain such info).
Rather than mutating the original exception's backtrace, this wraps it
with a proxy object.
Eventually we will implement backtrace_locations on the proxy object so
that the exception handling middleware can be updated to _only_ deal
with backtrace_locations and never deal with raw `backtrace`
We are currently mutating exception objects and I would like to stop
doing that. Unfortunately the views are calling many methods directly
on the exception and expecting that the mutations exist.
This patch refactors the templates so that they ask the ExceptionWrapper
class for information about the exception rather than directly asking
the exception object itself
`selenium-webdriver` v4.5.0 adds more entries ("acceptInsecureCerts" and
"moz:debuggerAddress") to the `as_json` output for
`Selenium::WebDriver::Firefox::Options`, causing an exact comparison of
the Hash to fail.
See SeleniumHQ/selenium@58f5833ba0.
Gotta be honest, this is so I can make some hacks. Basically I would
like an engine to specify where to find rescue templates, and currently
there's no way to add search paths to the debug view lookup context.
This commit turns the template path in to an array (that I plan to
mutate, but nobody should do that besides me until we make an actual
good API).
I added the `dup` in `initialize` so in case the array is accidentally
mutated we don't leak memory.
Previously, the method always asserts the status is `:redirect` which
allows for any kind of 3XX response. However, sometimes it is worthwhile
to precise the status code of the redirect. For example, a Rails
application may want to verify the redirect is a 301 (Moved Permanently)
and not the default 302 (Found). The new method argument makes this
convenient to do in one assertion.
Rubinius has not been maintained since May 2020 and based on the
discussion at https://github.com/rails/rails/pull/44984 ,
I think we can remove Rubinius specific code from Rails.
Previously ActionDispatch::ServerTiming would subscribe and unsubscribe
on each request. This could cause issues with the internal stacks of
ActiveSupport::Notifications, particlularly under the previous AS::N
implementation which used thread-local stacks for every subscriber
(the new implementation has mostly mitigated this).
Additionally, the previous ServerTiming implementation did not report
metrics correctly in a multi-threaded environment.
This commit works around both of these issues by using a single global
subscription, which collects events into a per-thread Array.
Without this change if action_dispatch.cookies_serializer is set to
json and the app tries to read a marshal-serialized cookie, it will
raise a JSON::ParserError which won't clear the cookie and force app
users to manually clear the cookie in their browser.
(See #45127 for original bug discussion)
This change incorporates to Rails a feature called error_highlight that
has been available since Ruby 3.1. This allow Rails' error report screen
to display the fine-grained location where an error occurred (not only a
line number but also beginning and end column numbers of the code
fragment).
For ErrorHighlight, see https://bugs.ruby-lang.org/issues/17930 in
detail.
To use error_highlight, ExceptionWrapper now prefers
`Exception#backtrace_locations` (since Ruby 2.1), which returns an array
of `Thread::Backtrace::Location`s, instead of `Exception#backtrace`.
This is because error_highlight requires `Thread::Backtrace::Location`
to locate the column where an error was raised.
Co-Authored-By: John Hawthorn <john@hawthorn.email>
Co-Authored-By: Jean Boussier <jean.boussier@gmail.com>
The various LogSubscriber subclasses tend to subscribe to events
but then end up doing nothing if the log level is high enough.
But even if we end up not logging, we have to go through the
entire notification path, record timing etc.
By allowing subscribers to dynamically bail out early, we can
save a lot of work if all subscribers are silenced.
