Collection>sum fails on empty collection

#() sum -> errorEmptyCollection.

#sum uses #reduce: and I can see why #reduce might consider an empty collection an error - but I also see that #sum of an empty collection probably ought to be 0, at least in a lot of cases. And I'd note that using a simple #inject:into: - in the case of a million item Array, at least - took about 60% of the time of the current #sum.

Why did this come up? In an MVC Project I tested
  PluggableFileList getFolderDialog openLabel: 'foo'
whilst seeing if PluggableFileList had any uses. It turns out that building a PluggableFileList in MVC doesn't work. Somewhere we end up trying to build a pluggablepanel with no children, which makes the logic in MVCToolBuilder>>setLayout:in: get upset. I don't care enough about making MVC stuff work to be willing to spend time on that.

tim
--
tim Rowledge; [hidden email]; http://www.rowledge.org/tim
Fractured Idiom:- RIGOR MORRIS - The cat is dead

> On 2019-01-08, at 12:10 PM, Eliot Miranda <[hidden email]> wrote:
>
> IIRC we discussed this issue carefully.  The issue was typing.  Clearly one should only be able to compute the sum of a collection of numbers (or things that answer to + and can be an argument to 0 + ...).  So we wanted to check that there was at least one numbers element in the receiver and hence chose that sum should indeed fail for an empty collection.  It still seems like a reasonable choice to me.  see e.g. http://forum.world.st/squeak-dev-Re-Collection-gt-gt-sum-implementation-td77002.html#a77008

I'm sure much thought was put into it; it's never easy trying to make really well crafted high-generality code. I certainly see the value for the #reduce: code as a general thing. I always did have a problem with messages like sum, average, median, and so on being implemented for general collections, for exactly the reason you point out - collections aren't all that often of purely numeric things.

In the particular case I stumbled over, the usage of #sum carries way too many assumptions and provides no way to recover from them. Indeed the entire algorithm in that part of the code collapses with a big "oh bugger".

tim
--
tim Rowledge; [hidden email]; http://www.rowledge.org/tim
Useful Latin Phrases:- Sentio aliquos togatos contra me conspirare = I think some people in togas are plotting against me.

> On 2019-01-08, at 1:54 PM, Nicolas Cellier <[hidden email]> wrote:
>
> You can always write:
>
>     ^collec inject: 0 into: #+

I'd not spotted that particular neat trick before - like it. It's a really interesting example of Smalltalk cleverness to make a simple symbol work as a block. Nice!

tim
--
tim Rowledge; [hidden email]; http://www.rowledge.org/tim
Multitasking:  Screwing up several things at once...

Use #detectSum::

#() detectSum: #yourself. "==> 0"

It also gives you full control:

#(true false 1 2 3.5 'foo') detectSum: [ :each |
  each isNumber "Yes, this won't work for complex numbers and quaternions"
  ifTrue: [ each ]
  ifFalse: [ 0 ] ]. "==> 6.5"


Levente

On Tue, 8 Jan 2019, tim Rowledge wrote:

> On 2019-01-08, at 2:26 PM, Levente Uzonyi <[hidden email]> wrote:
>
> Use #detectSum::

Also cool and previously unknown to me. Neat!

tim
--
tim Rowledge; [hidden email]; http://www.rowledge.org/tim
Strange OpCodes: RDRI: Rotate Disk Right Immediate