It's fundamentally flawed: the whole point of building in Docker is to
build in a hermetic environment. You can't mix building some parts in
Docker and some parts not. We'll build it only in Docker when making
release images, never depending on the host configuration.
Change-Id: I5fb283b17489fba18e89398e0d3f425c7f6159cd
From 13dfd8: 'Advertise 'h2' protocol, now that HTTP/2 is final.
We still advertise h2-14 too for a bit.'
Change-Id: Ic784937bf43c132c8a9259c50933839bedaef81a
First collect the embedable files, and check z_data.go's ModTime,
and whether it is empty.
This is not an error fix, nor a new feature, just an attempt to
eliminate a todo in make.go.
Change-Id: I0d20329ffb02000e56c9501cea87abc9e48b9f0e
fixes OOM errors, as reported in #414
Thanks to @hullerob for the heads up.
Also remove the test files because most of the tests are useless
without the sample data, which we don't ship.
Change-Id: Ia56b0184111d32a3c089785bf42cebb9b47af406
The problem was newSorted was taking care of doing the DB
initialization. It was not only unnecessary, since newKeyValueFromConfig
from sorted/sqlite will do it by calling initDB, but also incorrect
because WAL was not getting enabled. Which is what was causing the
"database is locked" error "fixed" in commit
f6bf779cd6
Change-Id: Ib4badeda4ed4c6c4be43d5f7fc4c543a467b0cd1
testSubFetcher in blobserver/storagetest was already checking that we'd
get specific error messages in the case of negative input parameters or
an out of range offset.
This change rationalizes these constraints with named errors
(ErrNegativeSubFetch and ErrOutOfRangeOffsetSubFetch) specified
in the SubFetcher interface.
It also fixes the googlestorage and s3 implementations so that they pass
the aforementioned test.
Change-Id: I25b72b842855b90ee3cab44c90654581dccf4b8e
this change introduces one API to filewriter, which takes an additional param for the file's lastmodtime to store with the file blob
Issue #317
Change-Id: I8da3134306c28e6945661296aaafb7752b4bf8b5
Reasons:
1) TestWebsocketQuery test now needs its world to have a corpus for its
query to succeed, because we now always sort unless specified
otherwise (and the query in the test does not do so).
2) it's more realistic to have an indexer with a corpus anyway, since
that's the canonical use-case.
Change-Id: Icdb31f8600f246227e237b851a1be825dc230d61
Fix some UI queries, since UI always queries with "-created".
Also add Unsorted, to override the CreatedDesc default, for when we
really do not want to sort, e.g. when we don't have a corpus and
trying to sort would make us error out.
Issue #579
Change-Id: Ife0aa816c5e8cac8dc6612d0ffd104238abc6838
So when you describe a file, you also gets its wholeref.
TODO: we'll need to migrate old indexes to this new format on
start-up.
Change-Id: I4a3fb000d68bde46474275c2070ef285a6d6ecfc
Brad Fitzpatrick