[R] Bioconductor / AnnotationDbi: Why does a GOAllFrame contain more rows than its argument GoFrame?

[Asis Hallab]

Dear Bioconductor Experts,

thank you for providing such a useful tool-set.

I have a question regarding the package AnnotationDbi, specifically
the classes GOFrame and GOALLFrame.

During a GO Enrichment Analysis I create a data frame with Arabidopsis
thaliana GO annotations and from that first a GOFrame and than from
this GOFrame a GOALLFrame. Checking the result with

nrow(  getGOFrameData(  athal.go.all.frame ) ) # The GOAllFrame

and comparing it with

nrow( athal.go.frame )                                     # The GoFrame

I realize that the GOALLFrame has more than 5 times more rows than my
original GO annotation table. If I provide
organism='Arabidopsis thaliana'
to the constructor of GOFrame this ratio increases even further.

Unfortunately I could not find any documentation on this, so I feel
forced to bother you with my questions:

1) Why does GOALLFrame so many more annotations?
2) Why and from where does it retrieve the organism specific ones that
are added when a model organism like 'Arabidopsis thaliana' is
provided?
3) I suspected that all ancestors of annotated terms are added, but
when I did so myself, I still got less GO term annotations? So do you
add ancestors of the "is_a" type and possibly other relationship types
like "part_of" etc. ?

Please let me know your answers soon. Your help will be much appreciated.

Kind regards!