read-depend branch is ready for another review...

[Robert Goldman]

On 7/17/17 Jul 17 -9:40 AM, Faré wrote:
> On Mon, Jul 17, 2017 at 9:47 AM, Robert Goldman <rpgoldman at sift.net> wrote:
>> On 7/16/17 Jul 16 -10:31 PM, Faré wrote:
>>>>> 1. what's in FIND-SYSTEM is accessors to %ADDITIONAL-INPUT-FILES, not
>>>>> ADDITIONAL-INPUT-FILES and the former has setters as well as getters, so
>>>>> I wasn't sure I could use ADDITIONAL-INPUT-FILES.  I'll check.
>>>> Checked and confirmed this -- %ADDITIONAL-INPUT-FILES takes only a
>>>> COMPONENT as argument, so it can't use the convenience methods macro.
>>>>
>>> Why do you need convenience methods for the internal accessor
>>> %ADDITIONAL-INPUT-FILES ? Can't users who need convenience methods
>>> (including internal functions) just use ADDITIONAL-INPUT-FILES, and
>>> let *its* convenience methods do the trick?
>> I was just being lazy, but I definitely found that the raw accessor was
>> being called with strings.  Our code structure doesn't make it easy to
>> be tidy about this.
>>
> The function didn't exist before, you are introducing it. Why do you
> call it with strings? Shouldn't this function always be hidden behind
> a call to ADDITIONAL-INPUT-FILES, that does have convenience
> methods???
> 

The call chain that led to my problem was as follows:

PARSE-DEFSYSTEM >
  PARSE-COMPONENT-FORM >
    NORMALIZE-VERSION  -- called with COMPONENT (!) a string
       RECORD-ADDITIONAL-SYSTEM-INPUT-FILE  -- component a string
            %ADDITIONAL-INPUT-FILES ---> FAIL


I've fixed that now...  As soon as I've finished all the tests, I will
push the patch.  Then we can merge.  I believe we can squash merge to
simplify the history, unless you think we need it...

Best,
r