freenode/#clasp - IRC Chatlog
Search
7:26:22
scymtym
currently, INTERPRET-TOKEN receives an array of Booleans indicating whether a given token character was escaped
7:26:54
scymtym
this is insufficient for distinguishing .. from .||. (the former is illegal, the latter is a symbol)
7:28:13
scymtym
what i would like to do, and implemented prototypically, is passing a list of escaped ranges, which can be potentially empty
7:29:29
scymtym
this is also more efficient for a couple of things since in the common case, the current escape indicator array is all false
7:30:35
scymtym
this also allows CONVERT-ACCORDING-TO-READTABLE-CASE to take a shortcut instead of looking at each character if there are no escapes. again, the common case
12:27:56
frgo
Yeah we have a bit of a heated discussion going due to some statements POTUS has made towards Germany ;-)
12:42:17
frgo
Hi beach. Well, you sort of have to decide how to handle him. Love him, ignore him, or change him.
12:42:59
drmeister
Yeah - he's an obnoxious asshole that the republicans are using to distract us while they reverse 50 years of progress here in the states, and shift us to an oligarchy.
12:44:01
beach
frgo: It seems like, as a result of some of his behavior, some European countries have decided that the appropriate response it to create a common defense organization.
13:20:47
Bike
i guess not that weird since uaets are done at that point so the switch is faster, but still
15:36:50
beach
And we need to determine whether to have data with associated types or have the types implied by the instructions.
15:37:19
Bike
https://eli.thegreenplace.net/2014/sfinae-and-enable_if/ how is C++ this absurd. i want an ifdef, i get... this
15:37:41
beach
drmeister: I am open to suggestions. I mean not just "I vote for X", but that come with some analysis of pros and cons.
15:38:21
Bike
the particular reason for the types is so that LLVM knows what sizes to use for stack allocations.
15:39:21
beach
I see. But that's a biased argument in my opinion. Though it is entirely possible that it is the best solution anyway, of course.
15:40:19
beach
There is also the question of how much to cater to similarity of existing processor, possibly closing the door to more unusual ones.
15:41:00
beach
That sounds very reasonable. I just want to hear some more arguments before making a decision.
15:42:12
Bike
sizes also matter with whatever aref-instruction would lower toin mir. i believe the "memref" instruction says it works with word addressing, the reasonable assumption being that an object pointer will be word sized
15:43:24
beach
I guess it is safe to make instructions that assume that, and then have client code use a different set of instructions for other architectures.
15:43:57
beach
We can just label certain instructions as being usable only on byte-addressed processors. I don't have a problem with that.