[#119390] [Ruby master Feature#20775] Gemify win32-registry, win32-sspi and win32-resolv — "larskanis (Lars Kanis) via ruby-core" <ruby-core@...>

Issue #20775 has been reported by larskanis (Lars Kanis).

12 messages 2024/10/01

[#119410] [Ruby master Feature#20778] ruby/repl_type_completor as a bundled gem — "tompng (tomoya ishida) via ruby-core" <ruby-core@...>

SXNzdWUgIzIwNzc4IGhhcyBiZWVuIHJlcG9ydGVkIGJ5IHRvbXBuZyAodG9tb3lhIGlzaGlkYSku

7 messages 2024/10/02

[#119432] [Ruby master Misc#20781] DevMeeting-2024-11-07 — "mame (Yusuke Endoh) via ruby-core" <ruby-core@...>

Issue #20781 has been reported by mame (Yusuke Endoh).

11 messages 2024/10/03

[#119442] [Ruby master Feature#20782] Introduction of Happy Eyeballs Version 2 (RFC8305) in TCPSocket.new — "shioimm (Misaki Shioi) via ruby-core" <ruby-core@...>

SXNzdWUgIzIwNzgyIGhhcyBiZWVuIHJlcG9ydGVkIGJ5IHNoaW9pbW0gKE1pc2FraSBTaGlvaSku

12 messages 2024/10/04

[#119462] [Ruby master Bug#20785] Should `a in b, and c` `a in b, or c` `a in b, rescue c` be syntax ok? — "tompng (tomoya ishida) via ruby-core" <ruby-core@...>

Issue #20785 has been reported by tompng (tomoya ishida).

10 messages 2024/10/05

[#119495] [Ruby master Feature#20792] String#forcible_encoding? — "kddnewton (Kevin Newton) via ruby-core" <ruby-core@...>

Issue #20792 has been reported by kddnewton (Kevin Newton).

16 messages 2024/10/09

[#119514] [Ruby master Bug#20796] Segmentation fault in rubyzip tests with ruby 3.4.0-preview2 — "tikkss (Tsutomu Katsube) via ruby-core" <ruby-core@...>

Issue #20796 has been reported by tikkss (Tsutomu Katsube).

10 messages 2024/10/13

[#119534] [Ruby master Bug#20800] Don't place `ruby` executable into `/usr/libexec/x86_64-linux/bin` — "vo.x (Vit Ondruch) via ruby-core" <ruby-core@...>

Issue #20800 has been reported by vo.x (Vit Ondruch).

10 messages 2024/10/21

[#119575] [Ruby master Bug#20808] Cannot override Data#inspect — "maicolben (Maicol Bentancor) via ruby-core" <ruby-core@...>

Issue #20808 has been reported by maicolben (Maicol Bentancor).

8 messages 2024/10/21

[#119621] [Ruby master Bug#20816] Potential regression in Ruby 3.3.x (compared with 3.1 and 3.2) regarding fast syscalls and multi-threading. — "adrienjarthon (Adrien Jarthon) via ruby-core" <ruby-core@...>

SXNzdWUgIzIwODE2IGhhcyBiZWVuIHJlcG9ydGVkIGJ5IGFkcmllbmphcnRob24gKEFkcmllbiBK

6 messages 2024/10/25

[#119622] [Ruby master Bug#20817] Ruby 3.4.0dev emits `warning: possibly useless use of + in void context` while Ruby 3.3.5 does not — "yahonda (Yasuo Honda) via ruby-core" <ruby-core@...>

Issue #20817 has been reported by yahonda (Yasuo Honda).

8 messages 2024/10/26

[#119646] [Ruby master Feature#20855] Introduce `Fiber::Scheduler#blocking_region` to avoid stalling the event loop. — "ioquatix (Samuel Williams) via ruby-core" <ruby-core@...>

Issue #20855 has been reported by ioquatix (Samuel Williams).

10 messages 2024/10/31

[#119650] [Ruby master Bug#20857] Don't change `Hash#inspect formatting` — "vo.x (Vit Ondruch) via ruby-core" <ruby-core@...>

Issue #20857 has been reported by vo.x (Vit Ondruch).

17 messages 2024/10/31

[ruby-core:119431] [Ruby master Feature#20669] Add Marshal::MarshalError class to differentiate ArgumentErrors

From: "olleolleolle (Olle Jonsson) via ruby-core" <ruby-core@...>
Date: 2024-10-03 09:46:13 UTC
List: ruby-core #119431
Issue #20669 has been updated by olleolleolle (Olle Jonsson).


In the meantime, Dalli merged a change that sounds a lot like Eregon's comment. https://github.com/petergoldstein/dalli/pull/1011

My immediate needs are met, and I will close this feature request.

----------------------------------------
Feature #20669: Add Marshal::MarshalError class to differentiate ArgumentErrors
https://bugs.ruby-lang.org/issues/20669#change-110046

* Author: olleolleolle (Olle Jonsson)
* Status: Open
----------------------------------------
Make it possible to differentiate general non-marshal errors from failures to decode Marshal data using a specific exception class.

## Background

There are a variety of error conditions that can cause Marshal to fail, including both corrupt data, and between versions of Ruby/Marshal binary formats, e.g.

```
> Marshal.load("foobar")
<internal:marshal>:34:in `load': incompatible marshal file format (can't be read) (TypeError)

> Marshal.load(Marshal.dump(Object.new).slice(0, 10))
<internal:marshal>:34:in `load': marshal data too short (ArgumentError)

> MyThing = Struct.new(:name, :age)
=> MyThing
> Marshal.dump(MyThing.new("Alice", 20))
=> "\x04\bS:\fMyThing\a:\tnameI\"\nAlice\x06:\x06ET:\bagei\x19"

(in a separate session)
> Marshal.load "\x04\bS:\fMyThing\a:\tnameI\"\nAlice\x06:\x06ET:\bagei\x19"
<internal:marshal>:34:in `load': undefined class/module MyThing (ArgumentError)
```

When data is corrupt, or incompatible, Marshal may raise either `ArgumentError` or `TypeError` with various messages. Not all errors are from `marshal.c`.

Generally, `TypeError` and `ArgumentError` are used to indicate some semantic problem with the program, rather than a runtime issue with the data. For example, `ArgumentError` may be used to indicate less than the required number of arguments passed to a method, and `TypeError` may indicate than the wrong data type was given as an argument.

Dalli is a library for accessing memcached servers. By default it uses Marshal to store serialised Ruby objects. However, when updating major versions of Ruby, or application code, `Marshal.load` may start raising `TypeError` or `ArgumentError`. This class of error needs to be handled differently from "normal" `TypeError` and `ArgumentError`. As such, the only way to differentiate right now is to pattern match the exception message, [which is done here](https://github.com/petergoldstein/dalli/blob/1d4cbfc78e6470ad57a883801a783bf30890c400/lib/dalli/protocol/value_serializer.rb#L36-L72):

```ruby
      # TODO: Some of these error messages need to be validated.  It's not obvious
      # that all of them are actually generated by the invoked code
      # in current systems
      # rubocop:disable Layout/LineLength
      TYPE_ERR_REGEXP = %r{needs to have method `_load'|exception class/object expected|instance of IO needed|incompatible marshal file format}.freeze
      ARGUMENT_ERR_REGEXP = /undefined class|marshal data too short/.freeze
      NAME_ERR_STR = 'uninitialized constant'
      # rubocop:enable Layout/LineLength

      def retrieve(value, bitflags)
        serialized = (bitflags & FLAG_SERIALIZED) != 0
        serialized ? serializer.load(value) : value
      rescue TypeError => e
        filter_type_error(e)
      rescue ArgumentError => e
        filter_argument_error(e)
      rescue NameError => e
        filter_name_error(e)
      end

      def filter_type_error(err)
        raise err unless TYPE_ERR_REGEXP.match?(err.message)

        raise UnmarshalError, "Unable to unmarshal value: #{err.message}"
      end

      def filter_argument_error(err)
        raise err unless ARGUMENT_ERR_REGEXP.match?(err.message)

        raise UnmarshalError, "Unable to unmarshal value: #{err.message}"
      end

      def filter_name_error(err)
        raise err unless err.message.include?(NAME_ERR_STR)

        raise UnmarshalError, "Unable to unmarshal value: #{err.message}"
      end
```

Unfortunately, there have been several incidents where the complexity of the current behaviour has lead to outages in production.

## Proposal

Ideally, the above code could use a specific (one or more) error class for detecting failed `Marshal.load`.

```ruby
      def retrieve(value, bitflags)
        serialized = (bitflags & FLAG_SERIALIZED) != 0
        serialized ? serializer.load(value) : value
      rescue Marshal::LoadError => error
        raise UnmarshalError, "Unable to unmarshal value!"
      end
```

One difficulty is retaining backwards compatibility. We may be able to create a sub-class of `class LoadError < ArgumentError` but it fundamentally seems wrong to me (should really be a `RuntimeError`).

## See also

- [Dalli PR where another error message would be added](https://github.com/petergoldstein/dalli/pull/1008)
- https://github.com/search?q=Marshal.load+ArgumentError&type=code - Git code search showing some variations of the problem (`rescue ArgumentError` and similar).



-- 
https://bugs.ruby-lang.org/
 ______________________________________________
 ruby-core mailing list -- ruby-core@ml.ruby-lang.org
 To unsubscribe send an email to ruby-core-leave@ml.ruby-lang.org
 ruby-core info -- https://ml.ruby-lang.org/mailman3/lists/ruby-core.ml.ruby-lang.org/


In This Thread

Prev Next