Project

General

Profile

Bug #15543

rb_str_set_len should clear code range

Added by nirvdrum (Kevin Menard) about 1 month ago. Updated about 1 month ago.

Status:
Rejected
Priority:
Normal
Assignee:
-
Target version:
-
ruby -v:
ruby 2.6.0p0 (2018-12-25 revision 66547) [x86_64-linux]
[ruby-core:91129]

Description

Calling rb_str_set_len on a String could alter the code range. I think this hasn't been much of an issue because of pure luck rather than anything that was deliberately designed. If called on a string that already has a CR_UNKNOWN code range, there's no problem because the correct values will lazily calculated.

An example invocation that could be problematic, using helper methods for writing C API specs from the Ruby Spec Suite, looks like:

@str = "abcdefghij"[0..-1]
@s.rb_str_set_len(@str, 1)
@str.should == "a"

@str.force_encoding(Encoding::UTF_8)
@str.valid_encoding?.should == true
@s.RSTRING_PTR_set(@str, 1, 'B'.ord)
@s.RSTRING_PTR_set(@str, 2, 0x80)
@s.rb_str_set_len(@str, 3)

p @str
@str.valid_encoding?.should == false # This line fails because the cached code range hasn't been updated.

The first call to valid_encoding? forces the code range to be calculated for the string and it's determined to be CR_7BIT. Then rb_str_set_len is called, simulating the splitting the bytes of a valid UTF-8 multi-byte character. Here, the code range is still cached as CR_7BIT, but the byte sequence is invalid for UTF-8.

I think the fix is a simple matter of clearing the code range in rb_str_set_len, but I'd appreciate a review of my analysis.

Associated revisions

Revision 7a86165d
Added by nobu (Nobuyoshi Nakada) about 1 month ago

[DOC] mention rb_str_modify

[ruby-core:91134] [Bug #15543]

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@66861 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

Revision 66861
Added by nobu (Nobuyoshi Nakada) about 1 month ago

[DOC] mention rb_str_modify

[ruby-core:91134] [Bug #15543]

History

Updated by nobu (Nobuyoshi Nakada) about 1 month ago

  • Status changed from Open to Rejected

It isn't an issue of rb_str_set_len.
The string isn't valid encoding by @s.RSTRING_PTR_set(@str, 2, 0x80).
You should call rb_str_modify before modifying a string object.

Updated by Eregon (Benoit Daloze) about 1 month ago

I think few people know about this, and rb_str_modify isn't documented in doc/extension.rdoc.
I guess some C extensions ignore this requirement and could have bugs because of it.

Isn't it simpler and safer to clear the coderange, as it's done for, e.g., rb_str_resize() ?

Updated by nobu (Nobuyoshi Nakada) about 1 month ago

rb_str_set_len is not the true problem.
Should RSTRING_PTR make the object unshared and clear the code range?
Or enclose it only for the core and prohibit in extension libraries?

Hmmm, the last choice may be the best, I think.

Updated by Eregon (Benoit Daloze) about 1 month ago

nobu (Nobuyoshi Nakada) wrote:

rb_str_set_len is not the true problem.
Should RSTRING_PTR make the object unshared and clear the code range?

That sounds safer, because indeed as soon as the C code can access the char* it can change it.

Or enclose it only for the core and prohibit in extension libraries?

I'm not sure what you mean?

Do you mean removing rb_str_set_len() or RSTRING_PTR from the public C-API?
Or do you mean in core, a special version of RSTRING_PTR would be used?
Or something else?

Also available in: Atom PDF