Bug #11661: sprintf causes a KeyError instead of using a default value for hash substitution

• Status: Closed
• Assignee: matz (Yukihiro Matsumoto)
• Target version: -
• Backport: 2.0.0: REQUIRED, 2.1: REQUIRED, 2.2: REQUIRED

[eddyluten (Eddy Luten)]

When using a format string that substitutes hash values (or using the % operator on a string), instead of using the Hash's default value if a key is not present, it causes a KeyError.

As an end-user, to get around this, my hash needs to know about all the possible keys ahead of time and pre-assign a value to them or handle the KeyError. Logically, I would assume that the sprintf implementation would use the default Hash value.

I wanted to open this issue to see what your collective thoughts were on this since I have a fork running locally that fixes this issue and was wondering if I could send a patch/PR for this.

This issue is reproducible using the following code:

my_hash = Hash.new('world')
puts "hello %{location}" % my_hash # expecting "hello world"
# "KeyError: key{location} not found"

[nobu (Nobuyoshi Nakada)]

• Tracker changed from Bug to Feature
• Status changed from Open to Feedback

GH-1087 doesn't work properly with a default proc expecting an argument, but segfaults.

My patch.

diff --git i/sprintf.c w/sprintf.c
index 705d3ce..eee5695 100644
--- i/sprintf.c
+++ w/sprintf.c
@@ -605,13 +605,10 @@ rb_str_format(int argc, const VALUE *argv, VALUE fmt)
 		}
 		CHECKNAMEARG(start, len, enc);
 		get_hash(&hash, argc, argv);
-		sym = rb_check_symbol_cstr(start + 1,
-					   len - 2 /* without parenthesis */,
-					   enc);
-		if (sym != Qnil) nextvalue = rb_hash_lookup2(hash, sym, Qundef);
-		if (nextvalue == Qundef) {
-		    rb_enc_raise(enc, rb_eKeyError, "key%.*s not found", len, start);
-		}
+		sym = rb_cstr_intern(start + 1,
+				     len - 2 /* without parenthesis */,
+				     enc);
+		nextvalue = rb_hash_aref(hash, sym);
 		if (term == '}') goto format_s;
 		p++;
 		goto retry;

[nobu (Nobuyoshi Nakada)]

• Status changed from Feedback to Assigned

[nobu (Nobuyoshi Nakada)]

• Tracker changed from Feature to Bug
• Backport set to 2.0.0: REQUIRED, 2.1: REQUIRED, 2.2: REQUIRED

[Hanmac (Hans Mackowiak)]

hm might it be possible that sprintf only uses default value/or default proc if the hash does has default,
but does still raise KeyError if it doesnt?

another idea would be that is does try to use hash[key] function if hash is a non-hash object.
(but that might reduce the calcing speed if not checked right)

[eddyluten (Eddy Luten)]

Hans Mackowiak wrote:

hm might it be possible that sprintf only uses default value/or default proc if the hash does has default,
but does still raise KeyError if it doesnt?

another idea would be that is does try to use hash[key] function if hash is a non-hash object.
(but that might reduce the calcing speed if not checked right)

Nobuyoshi Nakada: this was the change I initially made myself but couldn't get it to work properly with hash substitutions. Maybe I did something wrong in my testing, I will attempt to use your changes, thanks! :)

Hans Mackowiak: so raise KeyError is the default is nil? I think that could work, since if you wanted empty strings as a default, you could simply set an empty string as default value. However, logically, since "#{nil}" (nil.to_s as well for that case) works and returns an empty string, the modulus operator version of it should work the same. What do you think?

Also a general question, how would I go about making the Rubyspec tests pass for this particular pull request? Do I open another PR for the tests in Rubyspec? If so, how do I link these two PRs to work together in the Travis CI build acceptance test?

[nobu (Nobuyoshi Nakada)]

Unfortunately, this change conflict with rubyspec.
How should we do:

1. backport the fix and remove the failed spec
2. add version guard and fix only the trunk
3. keep the current behavior

[nobu (Nobuyoshi Nakada)]

• Status changed from Assigned to Closed

Applied in changeset r52530.

---

sprintf.c: hash default value

• sprintf.c (rb_str_format): respect default value of a hash. no
  longer raises KeyError unless the default value of the hash is
  nil. [ruby-core:71354] [Bug #11661]

[usa (Usaku NAKAMURA)]

• Related to Bug #11677: 52530の変更により、sprintfの引数のhashにkeyが存在していてもvalueがnilのときにKeyErrorがでるようになってしまった。 added