When you put a record beyond the "end" of the database, in the fixed-
length record world, the intermediate records weren't being created
correctly, nor were they filled with the fill character.
From

When you put a record beyond the "end" of the database, in the fixed-
length record world, the intermediate records weren't being created
correctly, nor were they filled with the fill character.
From: "Chris G. Demetriou" <cgd@postgres.Berkeley.EDU>

show more ...

if DB_LOCK not set, pin page across calls and don't copy key/data pair

4.4BSD snapshot (revision 8.1); add 1993 to copyright

break modify into two parts, recno file modify and btree modify
break BTF flags up into btree/recno (B_, R_) categories

delete R_CURSORLOG from the interface (and the code)

add BTF_EOF flag, test it instead of bt_reof
clean up mmap and file pointer uses in recno interface

bcmp/bcopy/bzero -> memcmp/memmove/memset

index_t -> indx_t

move db.h to the end of the includes, make compat.h work better

clear the stack in the search routine; it's a lot simpler

it's an insertion if R_IAFTER or R_IBEFORE are set, regardless

key can be set by put, now, so no longer const *; change R_APPEND to
R_CUSORLOG, with minor changes in semantics
remove BTF_DELCRSR, since recno just decrements the cursor, add test for
BTF_SEQINIT i

key can be set by put, now, so no longer const *; change R_APPEND to
R_CUSORLOG, with minor changes in semantics
remove BTF_DELCRSR, since recno just decrements the cursor, add test for
BTF_SEQINIT if using R_CURSOR
add in memory recno trees; add R_SETCURSOR, and return the key

show more ...

using the wrong variable to count when entering empty records, so
only a single empty record was ever entered. Also, not inserting a record
unless nrec > bt_nrecs.

make R_IAFTER of recno 0 work

check user flags; redo cursor implementation; recno doesn't care about
exactness of search results; do R_NOOVERWRITE at user interface level;
don't do parent page increments here

Rev #2. Redo the btree code add recno's and checkpoint