util.filechunkiter has been using a chunk size of 64k for more than 10 years,
also in years where Moore's law still was a law. It is probably ok to bump it
now and perhaps get a slight win in some cases.
Also, largefiles have been using 128k for a long time. Specifying that size
multiple times (or forgetting to do it) seems a bit stupid. Decreasing it to
64k also seems unfortunate.
Thus, we will set the default chunksize to 128k and use the default everywhere.
'unexpected putlfile response: None' when an http error occurs is not very
helpful.
Instead, leave the handling of urllib2.HTTPError exceptions to other layers.
The home of 'Abort' is 'error' not 'util' however, a lot of code seems to be
confused about that and gives all the credit to 'util' instead of the
hardworking 'error'. In a spirit of equity, we break the cycle of injustice and
give back to 'error' the respect it deserves. And screw that 'util' poser.
For great justice.
Python 2.6 introduced the "except type as instance" syntax, replacing
the "except type, instance" syntax that came before. Python 3 dropped
support for the latter syntax. Since we no longer support Python 2.4 or
2.5, we have no need to continue supporting the "except type, instance".
This patch mass rewrites the exception syntax to be Python 2.6+ and
Python 3 compatible.
This patch was produced by running `2to3 -f except -w -n .`.
Before the hack would replace 'heads' with 'lheads' no matter where it occured
in a batch command string.
Instead we will use a regexp to more carefully only match the 'heads' commands.
We did read the exactly the right number of bytes from the response body. But
if the response came in chunked encoding then that meant that the HTTP layer
still hadn't read the last 0-sized chunk and expected the app layer to read
more data from the stream. The app layer was however happy and sent another
request which had to be sent on another HTTP connection while the old one was
lingering until some other event closed the connection.
Adding an extra read where we expect to hit the end of file makes the HTTP
connection ready for reuse. This thus plugs a real socket leak.
To distinguish HTTP from SSH we look at self's class, just like it is done in
putlfile.
Avoid the intermediate limitreader and filechunkiter between getlfile and
copyandhash - return the right protocol and put the complexity where it better
can be managed.
These names were found using Cython; I was completely puzzled until
I searched the rest of the tree. It's icky to mess with another
module's namespace, but ickier yet to do so without a comment :-)
When changesets referencing largefiles are pushed then the corresponding
largefiles will be pushed too - unless the target already has them. The client
will use statlfile to make sure it only sends largefiles that the target
doesn't have. The server would however on every statlfile check that the
content of the largefile had the expected hash. What should be cheap thus
became an expensive operation that trashed the disk and the cache.
Largefile hashes are already checked by putlfile before being stored on the
server. A server should thus be able to keep its largefile store free of
errors - even more than it can keep revlogs free of errors. Verification should
happen when running 'hg verify' locally on the server. Rehashing every
largefile on every remote stat is too expensive.
Clients will also stat lfiles before downloading them. When the server verified
the hash in stat it meant that it had to read the file twice to serve it.
With this change the server will assume its own hashes are ok without checking
them on every statlfile.
Some consequences of this change:
- in case of server side corruption the problem will be detected by the
existing check on the client side - not on server side
- clients that could upload an uncorrupted largefile when pushing will no
longer magically heal the server (and break hardlinks) - a client will now
only upload its uncorrupted files after the corrupted file has been removed
on the server side
- client side verify will no longer report corruption in files it doesn't have
(Issue3123 discussed related problems - and how they have been fixed.)
This change separates peer implementations from the repository implementation.
localpeer currently is a simple pass-through to localrepository, except for
legacy calls, which have already been removed from localpeer. This ensures that
the local client code only uses the most modern peer API when talking to local
repos.
Peers have a .local() method which returns either None or the underlying
localrepository (or descendant thereof). Repos have a .peer() method to return
a freshly constructed localpeer. The latter is used by hg.peer(), and also to
allow folks to pass either a peer or a repo to some generic helper methods.
We might want to get rid of .peer() eventually.
The only user of locallegacypeer is debugdiscovery, which uses it to pose as a
pre-setdiscovery client. But we decided to leave the old API defined in
locallegacypeer for clarity and maybe for other uses in the future.
It might be nice to actually define the peer API directly in peer.py as stub
methods. One problem there is, however, that localpeer implements
lock/addchangegroup, whereas the true remote peers implement unbundle.
It might be desireable to get rid of this distinction eventually.
This implements a part of issue 3386. It batches the request for the status of
all largefiles in the revisions that are about to be pushed into a single
request, instead of doing N separate requests.
In a real world test case, this change was verified to save 1,116 round-trips to
the server. It only requires a client-side change; it is backwards-compatible
with an older version of the server.
Let R be a repo served by an hg daemon on a machine with an empty largefiles
cache. Pushing a largefiles repo to R will result in a no-such-file-or-directory
OSError because putlfile will attempt to create a temporary file in
R/.hg/largefiles, which does not yet exist.
This patch also adds a regression test for this scenario.
Before, a tempfile was used to create a temp file was created with 600
permissions and the uploaded data was written into it. This file was
then *copied* to .hg/largefiles/<hash>.
We now simply use atomictempfile to write the data to a temp file with
the right permissions and then rename that into place.
This is consistent with the rest of Mercurial's code, mirroring the
try-finally-unlink structure elsewhere. Furthermore, it fixes the case where
largefiles throws an IOError on Windows when the temporary file is opened a
second time by copytocacheabsolute.
This patch creates the temporary file in the repo's largefiles store rather than
/tmp, which might be a different filesystem.
The original intent was that the largefiles would primarily be in the
repository, with the global cache being only that--a cache. The naming
conventions and actual intent have both strayed. In this first patch, the
naming conventions are switched to match the actual intent, as are the
configuration options.
This is mainly about keeping code under the 80-column limit with as
few backslashes as possible. I am deliberately not making any logic or
behaviour changes here and have restrained myself to a few "peephole"
refactorings.
- fix some ungrammatical/unclear/incorrect comments/docstrings
- rewrite some really unclear comments/docstrings
- make formatting/style more consistent with the rest of Mercurial
(lowercase without period unless it's really multiple sentences)
- wrap to 75 columns
- always say "largefile(s)", not "lfile(s)" (or "big files")
- one space between sentences, not two