Feature #9841: [RFC] use ihash for fstring implementation

• Status: Rejected
• Assignee: -
• Target version: -

[normalperson (Eric Wong)]

This results in ~100K[1] memory savings on start up (with RubyGems
loaded) with no user-visible changes.

While this is not the best example of potential ihash-related memory
savings, there is zero-visible impact related to the loss of hash
ordering from this change.

Per-fstring entry overhead is now 24 bytes per-entry on 64-bit,
half of the previous 48 bytes from st_table_entry_t.

[1] as measured by the time(1) command on eglibc 2.13-38+deb7u1
(Debian 7.0 on x86-64)

Related to Feature #9614

(note: I will have limited Internet access until next week,
free to take your time in reviewing and responding)

Also viewable via git:
http://bogomips.org/ruby.git/patch?id=61c27401ed66 (1/2)
http://bogomips.org/ruby.git/patch?id=67f47a356a7f (2/2)

[normalperson (Eric Wong)]

• File ihash-fstring-only.mbox ihash-fstring-only.mbox added

[normalperson (Eric Wong)]

[e-mail] wrote:

This results in ~100K[1] memory savings on start up (with RubyGems
loaded) with no user-visible changes.

Wait, ccan htable is even cheaper in terms of memory usage. It saves
~150K at startup with fstring measured using time(1) on Debian 64-bit
eglibc: http://bogomips.org/ruby.git/patch?id=bddbbfb261b59

[naruse (Yui NARUSE)]

• Target version deleted (2.2.0)

[normalperson (Eric Wong)]

• Status changed from Open to Rejected