Changeset: 517258bcef42 for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=517258bcef42
Modified Files:
monetdb5/valgrindInfo
monetdb5/void2oid
Branch: default
Log Message:
Remove junk
diffs (94 lines):
diff --git a/monetdb5/valgrindInfo b/monetdb5/valgrindInfo
deleted file mode 100644
--- a/monetdb5/valgrindInfo
+++ /dev/null
@@ -1,4 +0,0 @@
-
-How to use valgrind?
-
-valgrind --tool=cachegrind --trace-children=yes Mserver --set gdk_alloc_map=no
tst5020.mal
diff --git a/monetdb5/void2oid b/monetdb5/void2oid
deleted file mode 100644
--- a/monetdb5/void2oid
+++ /dev/null
@@ -1,80 +0,0 @@
-Dec 21 2006, challenge replace :void columns by :oid{dense,base=0@0}
-as part of the BIG BANG SQL Switchpoint cleanup.
-
-Rationale: the current :bat[:void,-] column is a mixture of type information,
-value sequences, and storage implementation. Let alone the naming
-confusion, which already lead to coining it :bat[:vid,-] instead.
-The :bat[:void,-] column can be freely mixed with :bat[:oid,-] in
-the kernel, and many places in the M5 code use multiple signatures to
-capture all eligable cases/combinations. Moreover, it complicates
-the optimizers, because the typeing information may change as a result
-of runtime properties.
-
-Goal: to simplify the MAL language semantics and given its stable
-implementation, it seems possible to clearly separate the type
-information from the underlying (or desirable properties).
-That is, the old :bat[:vid,-] is turned into :bat[:oid{dense,base=0@0},-].
-
-It leaves open the situation to allow for a :void column if you
-really want a unary table. It is a none-space consumptive column type.
-Thus, the object :bat[:void,:any] has a constant header nil:void.
-You can append(-)/insert(nil:void,-) elements.
-It also enables positional lookup and positional updates.
-Such objects can be coerced into a :bat[:oid,:any]
-using a markH().
-
-The degenerated case is :bat[:void,:void], which contains only
-pairs (nil:void,nil:void). If we support bat.setNotNil() then
-it represents an always empty BAT.
-
-A consequence of this move could be that some runtime errors may
-emerge, because not all kernel wrapper code can deal with
-real-void headers yet.
-
-Moreover, it now becomes an issue when a physical :vid column is coerced
-into a physical :oid column under the hood. A call to bat.new(:oid,-)
-creates a :vid headed BAT starting at 0@0, which is coerced to a materialized
-:oid column if the sequences gets broken.
-
-Remapping the oids should be performed with a batcalc.+() operation.
-
-Analysis: a global analysis of the consequences is made before
-this drastic change is applied.
-
-Status:[28/12] all test cases dealing with :bat[:void,-] have been
-checked and recast into a :bat[:oid,-] version. The wrapper
-code still should partly be checked. A call to BATnew(TYPE_void,-)
-is only allowed if also the sequence base is set. This
-creates the :vid column. Ideally, the GDK library should make
-this type explicit (TYPE_vid) and distinct from TYPE_void.
-Now it is hidden in macros.
-
-src/modules/mal
- 40 cases where TYPE_void plays a decisive role
- 81 signatures are void specific[done]
-src/modules/kernel
- 136 cases where TYPE_void plays a decisive role
- 203 signatures are :void specific
- bat5.mx- 68 cases
- setSequenceBase() is dropped.
- update code for real-void headed bats should be handled.
- algebra.mx (30case)
- markT and friends involve oid-warping
- identity (dirty hack) is dropped
- aggr.mx
- most routines have to be protected against non-voids or masked
- they should be generalized anyway to avoid the strong limitation on oids
- batcalc.mx
- have to differentiate at runtime rather then compile time the algorithm
- and generalized to support oid columns
- group.mx
- generalize beyond void column
-src/mal
- 36 cases where TYPE_void plays a decisive role [check done]
-src/optimizer
- 20 cases where TYPE_void plays a decisive role [check done]
-*/Tests
- around 788 test files should be checked and re-approved
- a simple recast experiment in mal/Tests shows that only
- 23 out of 195 needs a re-approval when :void is replaced
- with :oid.
_______________________________________________
Checkin-list mailing list
[email protected]
http://mail.monetdb.org/mailman/listinfo/checkin-list
