Add flag to ruby to make warnings fatal.
I have a feature request: Please add a flag to the ruby command line to make warnings fatal.
It would be really handy when debugging and testing to make warnings fatal.
It's not always obvious what bit of code is causing the code that has the warning to execute. Having a full exception style trace back is very handy.
In addition, if you're doing tests, it would be good to have it raise an exception if a warning is fired. See http://stackoverflow.com/questions/660737/can-you-ask-ruby-to-treat-warnings-as-errors for an example.
#1 Updated by Aaron Patterson almost 5 years ago
I'm not sure this is necessary. For example, sprintf() will raise an exception if in debug mode:
irb(main):001:0> $-w = true
irb(main):002:0> "foo" % 10
(irb):2: warning: too many arguments for format string
irb(main):003:0> $DEBUG = true
irb(main):004:0> "foo" % 10
ArgumentError' at (irb):4 - too many arguments for format stringArgumentError' at /Users/apatterson/.local/lib/ruby/1.9.1/irb/workspace.rb:80 - too many arguments for format string
ArgumentError: too many arguments for format string
Maybe we should just update code to raise an exception when in debug mode like sprintf() does.
#4 Updated by Thomas Sawyer over 3 years ago
Hmm... I just wrote article related to this:
#5 Updated by Eric Hodel over 3 years ago
Sometimes when I use Kernel#warn in my code it is a message the user may not be able to do anything about, such as a deprecation message in library "a" that is used in library "b" that has not yet been updated, but the user wishes to use. Turning these into exceptions would break this use of warnings.
#7 Updated by Eric Hodel over 3 years ago
There are no priorities for warnings at present, so any code using warnings would need to be updated to take advantage of the feature, breaking backwards compatibility.
Existing uses of #warn could be set as "lowest priority" which makes the feature not very useful since only new code will opt in, and the exception will only be useful when ruby is run with -d, which is rare.
#8 Updated by Thomas Sawyer over 3 years ago
It wouldn't break backward compatibility, but it would take time for libraries to adjust to take the most advantage of it. And yet, any library with active development will likely adjust very quickly.
I would expect existing uses of warn to be set at "nominal priority" -- just below the error threshold. Although I imagine one could make the case that we should insert one priority layer between the two which would help make them a little more useful with libraries that haven't yet adjusted.