Comments on: Remember to breathe

By: Ted Gould

Ted Gould — Fri, 16 Oct 2009 19:58:52 +0000

Hmm, that didn’t work nearly as well as I hoped it would

By: Ted Gould

Ted Gould — Fri, 16 Oct 2009 19:52:18 +0000

Hey,

Used your Makefile.am.enum and ended up making a couple changes to make distcheck happy.

=== modified file ‘Makefile.am.enum’
— Makefile.am.enum 2009-10-16 17:59:39 +0000
+++ Makefile.am.enum 2009-10-16 19:50:10 +0000
@@ -24,10 +24,11 @@
EXTRA_DIST += $(enum_tmpl_h) $(enum_tmpl_c)

stamp-enum-types: $(glib_enum_headers)
+ $(QUIET_GEN)mkdir -p `dirname $(builddir)/$(glib_enum_h)`
$(QUIET_GEN)$(GLIB_MKENUMS) \
- –template $(enum_tmpl_h) \
+ –template $(srcdir)/$(enum_tmpl_h) \
$(glib_enum_headers) > xgen-eh \
- && (cmp -s xgen-eh $(glib_enum_h) || cp -f xgen-eh $(glib_enum_h)) \
+ && (cmp -s xgen-eh $(builddir)/$(glib_enum_h) || cp -f xgen-eh $(builddir)/$(glib_enum_h)) \
&& rm -f xgen-eh \
&& echo timestamp > $(@F)

@@ -35,9 +36,10 @@
@true

$(glib_enum_c): $(glib_enum_h)
+ $(QUIET_GEN)mkdir -p `dirname $(builddir)/$(glib_enum_c)`
$(QUIET_GEN)$(GLIB_MKENUMS) \
- –template $(enum_tmpl_c) \
+ –template $(srcdir)/$(enum_tmpl_c) \
$(glib_enum_headers) > xgen-ec \
- && cp -f xgen-ec $(glib_enum_c) \
+ && cp -f xgen-ec $(builddir)/$(glib_enum_c) \
&& rm -f xgen-ec

By: Jeff Walden

Jeff Walden — Thu, 15 Oct 2009 04:35:30 +0000

I wasn’t suggesting insufficiency at 32 bits; there are all sorts of natural quantities computers might measure which overflow 32 bits (wealth of several hundred people in the world in dollars, for example). If you change 32 to 64 in your example, tho, suddenly 43 days (I think you misplaced a decimal point) changes into five hundred million years. At that scale, who cares about the vanishingly rare twitpocalypse?

By: ebassi

ebassi — Thu, 15 Oct 2009 00:42:01 +0000

@Jeff: if you have a site with 10 millions of users who bang out at least 10 status messages per day you’ll overflow a signed 32 bits integer in 429 days. 10 status messages per user is a ridiculous number, if we start factoring in the celebs and the spammers. no wonder Twitter already went through a twitpocalypse and has spilled over to 64 bits. but, all in all, if you just want to assume a monotonic number you should be using a string to hold it, not use a native type that has no defined size and defaults to machine size. because we don’t have 128 bit architectures widely deployed on the entire planet (yet).

By: Jeff Walden

Jeff Walden — Wed, 14 Oct 2009 18:45:34 +0000

incremental integers for unique id without a known length limit? not good design.

Why so? How long would your program have to run, uninterrupted, to overflow 64 bits? Theoretically it might matter — but even the longest-running programs aren’t going to run anywhere near long enough to encounter those limits.

(This isn’t merely an idle question; I have one project that relies on unique-id integers for correctness, although as the project’s JS it’s really IEEE doubles which don’t lose precision until 2**52 or thereabouts. I figured it would take a few thousand years or so to overflow, so I didn’t bother worrying further.)