* master

- Fixed regression due to uppercase characters in kernel language
  names, e.g. R.  =emacs-jupyter= would not work at all with language
  specific features.

* v0.8.2

- Fix recursive loading error caused by =jupyter-tramp.el= autoloads.

- Consider kernelspecs with spaces and uppercase characters in the kernel's
  language name.

- In =org-mode=, fixed a bug where inline results where being displayed
  differently from what they would have been when the results are not inline.

- When =jupyter-repl-echo-eval-p= is non-nil, ensure that any cell code not sent
  to a kernel yet is preserved.

- In =org-mode=, fix a bug where if the =:session= header argument of Jupyter
  source blocks is a relative file name with a =.json= suffix
  and =default-directory= is a remote directory, a file on the local machine
  would be read.

- Use =org-babel-jupyter-async-inline-results-pending-indicator= as a placeholder
  when evaluating asynchronous inline source blocks in =org-mode=.

- Integrate better with packages like =polymode= when evaluating source blocks
  in =org-mode= by considering indirect buffers. Thanks to =@dvzubarev= for
  the [[https://github.com/dzop/emacs-jupyter/issues/171#issuecomment-522762377][solution]].

- Fixed usage of source block parameters when evaluating source blocks
  using =#+CALL= in =org-mode=. There are many sources of source block parameters
  that =org-mode= merges to produce the final set of parameters passed
  to =org-babel-execute:jupyter=, previously that function would just re-compute
  source block parameters. Now it uses the parameters it was given,
  see =org-babel-jupyter-current-src-block-params=.

- Add support for displaying evaluation results in the current buffer using
  overlays. See =jupyter-eval-use-overlays= and the relevant README sections.

- Fix =jupyter-repl-restart-kernel= to restart a kernel using a kernel manager
  when one is available. Previously, restarting a kernel using that function
  wouldn't work properly if the kernel was from a notebook server.

- Increase default timeout until a kernel is considered dead to 30 seconds when
  communicating directly with a kernel. More specifically, if a response from a
  kernel is not received after three consecutive 10 s periods on the heartbeat
  channel, the kernel is considered dead. Previously the timeout was five 1 s
  periods. This should fix =Kernel not alive= errors when using the REPL if the
  kernel being communicated with was on a slow connection.

- Ensure that a non-default value of =url-cookie-file= is handled properly in the
  notebook support.

- In =org-mode= handle ANSI control codes in stream output.

- In =org-mode= fix a regression where source block results were not being
  associated with =#+CALL= Org elements.

- Add support for the Jupyter Notebook REST API. See the relevant sections in
  the README.

* v0.8.1

- Add the commands =jupyter-org-(next|previous)-busy-src-block= which will jump
  to the next or previous busy source block in an =org-mode= buffer. Bound to =N=
  and =P= in the =Hydra= provided in =org-mode= (=C-c h=).

- Add customizable variable =jupyter-org-adjust-image-size= to avoid stretching
  images when =org-image-actual-width= is set so that it falls back to a default
  width for an inline image.

- Fix a message buffering issue when sending multiple requests to a kernel in
  quick succession, e.g. when calling =org-babel-execute-subtree= in =org-mode=
  when all Jupyter source blocks in the subtree are evaluated asynchronously.

- In =jupyter-org-insert-src-block= (1) adjust whitespace so that there is at
  least one blank line before and following an inserted source block (2) do not
  insert a new source block between a source block and its results, always
  insert the new block after a source block's results (3) do not move past an
  inlinetask, normally the current paragraph or element under =point= is moved
  past before inserting a new source block.

- Fix issues with undo in the REPL when trying to undo after inserting a
  newline. See #139.

- In =org-mode=, detect Org formatted table strings when a kernel returns a
  =text/org= mimetype. This avoids wrapping the table in a drawer.

- If a file is opened via =emacsclient=, attempt to associate the opened buffer
  with a REPL client. This is useful, e.g. when using the ~julia~ language's
  ~edit~ function and the =EDITOR= environment variable is set to
  =emacsclient=. See ~jupyter-server-mode-set-client~ for more details.

- In =org-mode=, when inserting the results of an asynchronously evaluated
  src-block, indent the results to match the surrounding indentation level.

- In =org-mode=, fix an issue when a src-block is nested within a list that
  would cause the entire list to be erased when asynchronously evaluating the
  src-block.

- Add new customizable variable ~jupyter-repl-echo-eval-p~ that causes the REPL
  to behave similarly to comint or ESS when evaluating code. Specifically, it
  causes any code that is sent to the underlying kernel to also be displayed as
  a REPL input cell. The previous behavior would only display the sent code as
  an input cell if it was explicitly typed in at the REPL.

- Add the new command ~jupyter-repl-clear-cells~ to clear all of the current
  input and output cells in a REPL buffer. Bound to ~C-c C-o~ by default in the
  REPL.

- Fix an issue where timeouts started using ~with-timeout~ would timeout when
  the kernel requests input from the user. The timeouts are now suspended in
  such cases.

- In =jupyter-R.el= handle the =isolated= metadata property of HTML results by
  opening up an external browser to display the result.

- Allow =:session= to be a remote file name for =org-mode= source blocks. When
  =:session= is a remote file name that doesn't end in =.json=, e.g.
  =/ssh:ec2:jl=, a new kernel is started on the remote host using the
  =jupyter kernel= command on that host. The local file name part serves to
  distinguish different sessions on the remote host.

- Add the new commands ~jupyter-repl-history-previous-matching~ and
  ~jupyter-repl-history-next-matching~, bound to ~C-c M-r~ and ~C-c M-s~ in the
  REPL.

- In ~jupyter-with-display-buffer~, ~let~ bind ~jupyter-current-client~. This
  fixes an issue where kernel language specific methods were not overriding the
  defaults when attempting to display the contents of =display-data= messages
  received from a kernel.

- Fix a regression in ~jupyter-available-kernelspecs~ when getting the
  kernelspecs on remote hosts.

- Handle empty result blocks when stripping ANSI escape sequences during export
  in ~org-babel-jupyter-strip-ansi-escapes~.

- Call ~downcase~ on the language name when adding to ~org-src-lang-modes~.
  This fixes an issue with the ~C++~ kernel where the corresponding mode is
  ~c++-mode~.

* v0.8.0

- Support Windows

- In =org-mode=, support a header argument =:pandoc t=. Whenever this header
  argument is provided, certain kinds of code block results are translated into
  =org-mode= syntax using a =pandoc= process, if available. See the variable
  =jupyter-org-pandoc-convertable= for the kinds of results which can be
  converted in this way.

- Add initial support for C++.

- Add a new customizable variable =jupyter-repl-allow-RET-when-busy= which
  allows =RET= to insert a newline in a REPL cell whenever the kernel of the
  REPL is busy. Normally this isn't allowed since, when a kernel is busy, it
  doesn't respond to messages such as the =:is-complete-request= which
  is used to determine when the code should be sent to the kernel or a newline
  should be inserted whenever =RET= is pressed.

- Add initial support for R. Thanks to =jackkamm=.

- In =jupyter-eval-string= also handle =:display-data= messages.

- Add a new =jupyter-comm-layer= class which generalizes how a
  =jupyter-kernel-client= communicates with a kernel. Previously there was only
  one method of communication with a kernel, via a subprocess which
  communicates to the kernel using ZMQ sockets. In order to support other kinds
  of communication, e.g. using websockets, the =jupyter-comm-layer= generalizes
  the method of communication so that a client need not know how communication
  occurs.

- Add =jupyter-org-execute-subtree= (=C-M-<return>=) to the =jupyter-org=
  =hydra=. Also, various improvements to the source block insertion functions
  such as how they handle whitespace and determine the language of the source
  block to insert.

- Properly cleanup connection files when restarting a kernel.

- Fontify ANSI coded regions in example blocks and fixed width elements in
  =org-mode=. This feature is activated when =jupyter-org-interaction-mode= and
  currently applies to all example blocks and fixed width elements. This is
  useful for kernels that apply ANSI color codes to source block results such
  as IJulia.

  - Note the ANSI codes will be removed from the elements during export so that
    they don't appear in the final exported document.

- In =org-mode= when =point= is on a Jupyter source block inside an invisible
  region, e.g. when it =point= lies inside a folded subtree, ignore any special
  keybindings for the source block.

- Consider the =/docker:= TRAMP method provided by
  https://github.com/emacs-pe/docker-tramp.el. See [[orgit-rev:.::4ef27c2][(magit-rev 4ef27c2)]] for more
  details.

- Change the default completion context to only consider the deepest nested
  parenthetical level. Previously all levels were considered, but this is not a
  good default when interacting with LISP like kernels such as Clojupyter.

- In =org-mode=, when evaluating Jupyter source blocks during export, ensure
  that all source blocks are evaluated synchronously even when =:async yes= is
  specified.

- When using =jupyter-eval-line-or-region= and friends, don't display a message
  in the =*Messages*= buffer if a traceback will be shown.

- Be consistent with the use of the =:file= header argument when =:async no= is
  specified vs =:async yes=. In particular, the =:file= header argument means
  to use the path specified as the name of any image file generated by the
  source block in =emacs-jupyter=. For regular =org-mode= source blocks, it
  means to write the *results* of the source block to file. See #40 and #59.

- Fix off by one line issue when =jupyter-org-toggle-latex= is non-nil.

- Don't wrap the source code of an IJulia source block in a =let= block when
  using the =:dir= header argument in =org-mode= source blocks.

* v0.7.3

- Fix an issue where loading =jupyter-org-client.el= would trigger a recursive
  require of =org=.

- Don't fail if a short result has =%= characters in the default
  =jupyter-eval-short-result-display-function=.

- Port scimax's source block helper functions. Thanks to =UndeadKernel= for the
  work he put into this. See #45.

- Fix an issue where a kernel would be stuck in the busy state waiting for
  input from the user when calling =jupyter-eval-line-or-string= and friends.
  The =:input-request= handler was being inhibited in these cases.

- When evaluating Julia code blocks with the =:dir= header argument, fix an
  issue where top-level expressions like =using Plots= wouldn't work see #57.

- Don't use proportional fonts when rendering HTML. See #52.

- Fix whitespace issues when wrapping =org-mode= source block results in a
  =RESULTS= drawer.

- Increase =jupyter-default-timeout= to 2.5 seconds. See #43.

- Fix issues with the REPL restart process. See #50.
  - Refactor REPL restart to avoid relying on a =status: starting= message to
    be received.

  - Actually restart the kernel process when using a kernel manager.

  - In =jupyter-repl-restart-kernel= inhibit message handlers from running when
    sending a shutdown request.

- Add the functions =org-babel-jupyter-override-src-block= and
  =org-babel-jupyter-restore-src-block= to provide a way of overriding normal
  =org-mode= source block languages to use the Jupyter source block framework.
  For example, by evaluating =(org-babel-jupyter-override-src-block "python")=
  all =python= source blocks will actually be =jupyter-python= source blocks.

- Add the customizable variable =jupyter-pop-up-frame= which controls if a
  frame or a window is popped up when evaluating code using
  =jupyter-eval-line-or-region= (bound to =C-c C-c= when
  =jupyter-repl-interaction-mode= is enabled)

- In =jupyter-repl-after-change=, maintain the text properties at the beginning
  of a REPL input cell when text is deleted at the beginning of a cell. See #38.

- In =jupyter-repl-font-lock-fontify-region= and
  =jupyter-repl-syntax-propertize-function= narrow to the REPL input cell
  before doing any work since the kernel language mode's fontification
  functions do not know about REPL input/output boundaries.

- In =jupyter-repl-do-after-change=, widen the buffer before doing any work to
  take into account changes that narrow to fields. In such cases functions like
  =jupyter-repl-cell-code-beginning-position= will not work right since they
  look at positions before the =cell-code= field. See #38.

- Handle the overflow of the prompt margin in the REPL by increasing the margin
  width and re-calculating all the prompts in the buffer when the length of the
  prompt string exceeds the width of the margin. See #39.
  - Also, for the Julia kernel, don't replace the REPL prompt but add it as part of
    the REPL cell.

- Add support for suppressing =org-mode= table output from source blocks by
  specifying =:results scalar=.

- Add a new =org-mode= source block header argument =:display= which allows a
  user to control which mimetype is displayed. See #17.

- Fix an issue with undo in the REPL where the addition of continuation prompts
  for multi-line input would add extra undo information that would interfere
  with undo.

- Update =org-mode= source block result insertion to consider changes in how
  =fixed-width= and =example-block= elements and are printed to the buffer in
  =org-mode= >= 9.2. Also start testing against the latest version of
  =org-mode= in Travis.

- Handle the case of an empty =RESULTS= drawer during source block result
  insertion in the =:async yes= case.

- Add a =file= slot to a =jupyter-org-request=. Also, internally remove the
  =:file= header argument from an =org-mode= source block's parameters during
  block evaluation so that =org-mode= doesn't specially handle the =:file=
  argument as it interferes with insertion of results when =:async yes= is
  specified. Note this is currently only done for the =:async yes= case.

- In =jupyter-repl-syntax-propertize-function=, in addition to handling
  parenthesis syntax, handle string syntax. This is so that any string syntax
  characters in the output are not considered strings in the kernel's language.

- In =jupyter--display-eval-result= prefer Markdown if it is available.

* v0.7.2

- In =jupyter-handle-input-request=, be more secure when reading passwords by
  using =clear-string= after sending the message and avoiding printing
  passwords to the =*Messages*= buffer when =jupyter--debug= is non-nil.

- In =jupyter-insert-latex= remove modification-hooks from the image overlays
  so that the images are not removed from the buffer when changing the text
  properties of the underlying text.

- In =jupyter-read-expression= show the kernel language when prompting in the
  minibuffer.

- In =jupyter-repl-kill-buffer-query-function= also ask to kill the kernel.

- In =jupyter-with-display-buffer= properly advance
  =jupyter-display-buffer-marker= when contents are added to the buffer so that
  future inserts will insert at the end of the buffer. Also handle a =nil=
  RESET argument.

- Keep =:= as the start of a completion prefix for the Julia kernel to allow
  completing dictionary keys.

- Consider the state of the ioloop in the =jupyter-channel-alive-p= method of a
  =jupyter-kernel-client=. In particular, ensure the method returns nil, when
  the ioloop isn't alive.

- Change the default completion context, the context returned by
  =jupyter-code-context= with an argument of =completion=, to return all nested
  levels of parenthesis.

- In =org-babel-jupyter-setup-export=, use the kernelspecs on the system to add
  to =org-latex-minted-langs=.

- Add new hook variables =jupyter-repl-cell-{pre|post}-send-hook= that are
  called before and after sending the contents of a REPL cell to the kernel.

- In =jupyter-repl-finalize-cell= ensure that only the last cell in the REPL
  buffer is finalized by going to =point-max= before accessing a cell's
  properties. This avoids issues with modifying the properties of previously
  finalized cells.

- Integrate more with =font-lock= and =syntax-ppss= in the REPL by (1) adding a
  custom =syntax-propertize-function= (2) handle REPL mode characters in the
  Julia kernel and (3) use the kernel language syntax table when calling
  =font-lock= functions.

* v0.7.1

- Fix a bug in =jupyter-connect-repl= which would cause the REPL to think that
  the kernel wasn't alive. The issue was that the heartbeat channel was not
  exchanging messages with the kernel and the heartbeat channel is relied on to
  check for the liveness of a kernel connected to using =jupyter-connect-repl=.
  See #29.

- When using =jupyter-eval-string= (=C-c M-:=), properly use the client local
  variable, =jupyter-eval-expression-history=, as the minibuffer history.

- =jupyter-repl-restart-kernel= now prompts for a REPL client to restart if the
  =jupyter-current-client= variable is not set in the current buffer. See #28.

- Fix bug when a kernel does not respond to a shutdown request. Previously in
  such cases, the kernel process would not be forcibly killed and would stay
  alive.

- Add =org-babel-jupyter-setup-export= to integrate the exporting process with
  =emacs-jupyter=. This function is added to
  =org-export-before-processing-hook= and currently only ensures that, when
  exporting to LaTeX and the minted package is being used, the =jupyter-LANG=
  source blocks use =LANG= for their minted language.

* v0.7.0

- Remove compatibility with =ob-ipython= by going back to using a =jupyter-=
  prefix instead of a =jupy-= prefix for Jupyter src-block languages.

- Re-use windows displaying =jupyter= specific buffers instead of popping up
  new windows whenever possible, e.g. when displaying a traceback or output
  caused by evaluating code. See =jupyter-display-current-buffer-reuse-window=.

- Consider the underlying REPL client of =org-mode= Jupyter src-blocks as valid
  clients to associate a source code buffer with using
  =jupyter-repl-associate-buffer=.

- Add the customizable variable =jupyter-org-toggle-latex= which automatically
  converts latex fragment results of =org-mode= Jupyter src-blocks into images
  if non-nil.

- Add the customizable variables =jupyter-eval-short-result-max-lines= and
  =jupyter-eval-short-result-display-function= which control how to display
  evaluation results having a number of lines less than
  =jupyter-eval-short-result-max-lines=. As an example, you can set
  =jupyter-eval-short-result-display-function= to =popup-tip= from the =popup=
  package to show short results inline after evaluation.

- When =:results silent= is an argument for an =org-mode= src-block and an
  error occurs, display a link to jump to the line of the src-block which
  caused the error along with the error traceback. Note this requires that the
  underlying kernel language extend the =jupyter-org-error-location= method.

- Fix integration with =insert-for-yank= inside a REPL buffer. Previously,
  yanking text from the kill ring into the REPL buffer would interfere with
  font-lock and motion functions such as =beginning-of-line=. See [[https://github.com/dzop/emacs-jupyter/issues/14][#14]].

- Add the minor mode =jupyter-org-interaction-mode= enabled in all =org-mode=
  buffers by default. This mode enables completion in Jupyter src-blocks
  directly from the =org-mode= buffer and custom keybindings for each kernel
  language that are only enabled if =point= is inside a Jupyter src-block. You
  bind keys to commands using =jupyter-org-define-key=. Inspired by =scimax=.

- Support the =:dir= header argument of =org-mode= src-blocks. Since Jupyter
  src-blocks have a backing REPL session, the =:dir= argument only ensures that
  the REPL session is initialized in the specified directory. After the session
  is initialized, the =:dir= argument has no effect when evaluating src-blocks
  with the same underlying session. Now, the directory is changed inside the
  REPL environment before evaluation of a src-block and reset to the previous
  directory after evaluation whenever =:dir= is specified as a header argument.
  Note, this requires that the backing kernel language handles =:dir= in the
  changelist argument of =org-babel-jupyter-transform-code=. Also inspired by
  =scimax=.

- Add support for inline Jupyter src-blocks in =org-mode=.

- For Jupyter src-blocks, delete files of unreachable links from
  =org-babel-jupyter-resource-directory=. When replacing image link results of
  a src-block, e.g. by re-evaluation of the src-block, delete the corresponding
  image file if it exists in =org-babel-jupyter-resource-directory=. Once again
  inspired by =scimax=.

- Add the =jupyter-repl-traceback= face. This face is used to fontify the
  background of a traceback in the REPL buffer to distinguish it from other
  output. In addition to this face, there is also =jupyter-repl-input-prompt=
  and =jupyter-repl-output-prompt=.