summaryrefslogtreecommitdiff
AgeCommit message (Collapse)Author
2021-07-13gofmtNick White
2021-07-13internal/pipeline: Reorganise interfaces so that functions only declare what ↵Nick White
they need We were using Pipeliner as a catch-all, but it's nicer if the functions can just state that e.g. they need download functionality, so decompose things so that that's how we do things now.
2021-07-13aws: Only look up test queue id when asked for, as for most Init()s it won't ↵Nick White
be needed
2021-07-12Add necessary pipeliner dependency for testqueue (probably remove this from ↵Nick White
internal library later as its only needed for tests
2021-07-12Add test for upAndQueue functionNick White
This involved adding a test queue, so it can be run safely without intefering with the pipeline.
2021-07-08rescribe: Exit with an error if directory doesn't existNick White
2021-06-29rescribe: add documentation on how to generate embedded dataNick White
2021-06-29rescribe: Add embed target for darwin (osx) tooNick White
2021-06-22rescribe: Remove erroneous unnecessary mkdirNick White
2021-06-22rescribe: Make it clearer that embedded training files are available to useNick White
2021-06-22rescribe: add embedded tesseract for linuxNick White
2021-06-22rescribe: allow use of embedded training even if -systess is usedNick White
2021-06-22cloud: update spot image to latest version that wont attempt to build ↵Nick White
rescribe tool
2021-06-22rescribe: Add go generate command to download the needed files to embedNick White
2021-06-22rescribe: Add an embedded tessdataNick White
2021-06-21Merge remote-tracking branch 'ssh/master'Nick White
2021-06-21rescribe: Set up so only Tesseract needed for the build platform is embeddedNick White
2021-06-21rescribe: Embed Tesseract into binary so that no Tesseract install is necessaryNick White
2021-06-21update spot image usedNick White
2021-06-15pipeline: Ignore hidden files when checking and uploadingNick White
This prevents issues if a .DS_Store file is present in a directory.
2021-05-31local: Only create a file once we are sure that it will be writeableNick White
2021-05-31Add a test for up(), and document download() and up() properlyNick White
2021-05-31Fix bug after changing pipeliner for tests, to ensure DeleteObjects is ↵Nick White
available to Pipeliner
2021-05-19Close process channel after writing to err channel in download(), in case of ↵Nick White
an error This is needed so that in tests the error can be selected out reliably, rather than an empty process signal.
2021-05-19Add tests for download()Nick White
2021-05-19Fix syntax with another Errorf callNick White
2021-05-19Local download now tries to open the source file before creating a ↵Nick White
destination file, so if it fails an empty file isnt left behind
2021-05-19Add basic DeleteObjects implementation to local.goNick White
2021-05-19Fix syntax for some fmt.Errorf callsNick White
2021-04-12Update preproc dependencyNick White
2021-03-16rescribe: change default training directory to trainings/v0.3.3Nick White
2021-02-22lspipeline: Rename to lspipeline-ng, and restore pre concurrency version to ↵Nick White
lspipeline as there are some hard to debug issues in concurrency version
2021-02-15getsamplepages: Add -prefix option, and use 'best' to get random page numbersNick White
The -prefix option is useful to us. Previously only a .jpg for page number 100 was retreived, which failed if the book had fewer (or unusually named) pages, and also didn't provide a corresponding .hocr at all (bug introduced with 48958d2). Using 'best', which is (effectively) randomly sorted, provides a guaranteed to exist page, and a random one at that.
2021-02-05Merge branch 'master' of ↵Nick White
ssh://ssh.phx.nearlyfreespeech.net/home/public/bookpipeline
2021-02-05Update go-chart dependencyNick White
2021-02-01Update AWS dependency to 1.37.1Nick White
2021-02-01Ensure DeleteObjects can handle over 1000 files to delete; fixes rmbook for ↵Nick White
large books
2021-01-26Make ListObjectsWithMeta generic again and create a specialised ↵Nick White
ListObjectWithMeta for single file listing, so we can still be as fast, but do not have a misleading api
2021-01-26Improve lspipeline concurrency by removing WaitGroup stuffNick White
2021-01-26Speed up lspipeline by making s3 requests concurrently and only processing ↵Nick White
single results from ListObjects requests
2021-01-26Stop limiting keys returned from listobjectprefixes' api usage; this speeds ↵Nick White
up the request markedly
2020-12-15[rmbook] Append / to end of bookname, to ensure e.g. "1" doesnt match all ↵Nick White
books starting with "1"
2020-12-15[rmbook] Add -dryrun flagNick White
2020-12-14Add rmbook toolNick White
2020-12-14Update preproc module used to incorporate an important crash fixNick White
2020-12-07[rescribe] Fix up *.hocr glob, which ensures that using a savedir that ↵v0.3.2Nick White
already has a hocr directory in it will work
2020-12-07[rescribe] Allow saving of results to somewhere other than a directory named ↵Nick White
after the book being processed
2020-12-04Ensure mkdir will succeed in uploadNick White
2020-12-03[rescribe] Fix portability issue where hocrs may not be correctly moved and ↵Nick White
txt-ified on windows
2020-12-03Don't upload binarised pdf twice needlesslyNick White
This can also result in the file being uploaded twice simultaneously, as up() is running in a separate goroutine. This can cause failures on Windows as the file is attempted to be removed by one upload process while being open to upload by the other process. Probably it could also fail if the process completed by one (so the file was deleted) before being started by the other.