cl+ssl::libcrypto is also read at startup. If not properly closed in the
hooks, together with libssl, libcrypto will be loaded at startup and if
the first cl+ssl file alternative fails, the debugger is invoked.
* Fixes#1370
The CAST should target `unsigned` rather than `integer`, so that we are compatible with MySQL 5.7. Also empty tables might have NULL entries, which we transform to zero entries here, as expected by the Lisp code.
Fixes#1127.
This should help optimise the duration of migrating databases with very big
tables and lots of smaller ones. It might be a little too naive as far as
the optimisation goes, while still being an improvement on the default
alphabetical one.
Fixes#1099.
As our API to connect to MS SQL does not provide a facility to set the
target port number, place it in the TDSPORT environment variable, which is
reported to have the expected impact.
Should fix#1094.
That's a blind fix, or almost blind: we're given some information but I'm
not in a position to test the fix myself. Hope it works.
Fixes#1036.
Fixes#1018.
Fixes#1082 where letters from the URI are doubled because of the
look-ahead in the previous parser's implementation.
Thanks @svantevonerichsen6906 for the fix!
See #1016 where we try to build the DDL for a foreign key that references
tables that are not found in our catalogs. We should probably just ignore
those foreign keys, as we might have a partial load to implement.
That's another blind fix, but it looks like this should be doing it.
Hopefully. Chances to break pgloader even more are quite slim, so let's try
it in the master branch.
Fixes#1069.
It was only implemented for Postgres sources even though the implementation
is generic enough to be shared. It's only a matter of instructing our parser
about the new facility, which this patch does.
Fixes#1062.
The pg_catalog.pg_attribute column adsrc is populated only once, and then
not maintained by the server. The function pg_get_expr(adbin, adrelid) is
what to use and has been here since release 8.0, so that's what pgloader
should be using.
Fixed#1934.
It used to be that our casting rules mechanism would allow for matching
unsigned data types only, and we sometimes have a need to do special
behavior on signed data types.
In particular, a signed bigint(20) in MySQL has the same values range as a
PostgreSQL bigint, so we don't need to target a numeric in that case. It's
only when the bigint is unsigned that we need to target a numeric.
In passing update some of the default casting rules documentation to match
the code.
Fix#982.
Implement a generic-function API to discover the source database schema and
populate pgloader internal version of the catalogs. Cut down three copies of about
the same code-path down to a single shared one, thanks to applying some amount
of OOP to the code.
The cl-db3 lib just got improvements for new dbase file types and field
types, reflect those in pgloader.
Also, cl-db3 now can read the encoding of the file (language driver)
directly in the header, meaning we can rely on that metadata by default, and
only override it when the users tells us to.
See #961.
Before this patch Travis would build the pgloader image twice, one when
doing the `make clones save` command and another time when doing the main
command `make check`, that depends on Make targets that `make save` does not
build, such as buildapp.
Also, fix `make check-saved` to stop requiring that we save pgloader again.
Finally, use uiop:featurep to clean-up the code in the src/save.lisp file.
That's to avoid a name conflict with the cl-db3 package. In SBCL 1.5.2 in
Travis this conflict is an hard error and provoque failure to build the
pgloader binary.
Before that it was necessary to install a function in the lisp environment
either in the source itself in src/utils/transforms.lisp, or in a lisp file
loaded with --load-lisp-file (or -l for shorts).
While this could be good enough, sometimes a very simple combination of
existing features is required to transform a function and so doing some
level of lisp coding directly in the load command is a nice to have.
Fixes#961.
Fixed file formats might contain an header line with column names and a hint
of the size of each column. While it might be a long shot that we can
acutally use that as a proper fixed-format specification, this patch
implements a guess mode that also outputs the parsed header.
In case when the parsing is wrong in some level of details, it might
actually be a good start to copy/paste from the command output and go from
there.
Fixes#958.
