My GitHub Pull Requests

While plugins should be ordered, this attribute seems to be used as a map from an extension to that list of plugins rather than the list itself; as indicated in the getPlugins() method.

In my site I don't do any configuration of this, just using the defaults. While trying to update to Eleventy 3 I was running into errors

Problem writing Eleventy templates:

1. Having trouble writing to "./public/css/base.css", "./public/css/base.css.map" from "./pages/styles.11ty.js" (via EleventyTemplateError)
2. Transform @11ty/eleventy/html-transformer encountered an error when transforming ./pages/styles.11ty.js. (via EleventyTransformError)
3. plugins is not iterable (via TypeError)

Original error stack trace: TypeError: plugins is not iterable at HtmlTransformer._getPosthtmlInstance (file:///home/ats/vc/personal/web/node_modules/.pnpm/github.com+11ty+eleventy@c39e8961a2ab7e311696d415fc4238e8fe4d4d80/node_modules/@11ty/eleventy/src/Util/HtmlTransformer.js:39:39) at HtmlTransformer.transformContent (file:///home/ats/vc/personal/web/node_modules/.pnpm/github.com+11ty+eleventy@c39e8961a2ab7e311696d415fc4238e8fe4d4d80/node_modules/@11ty/eleventy/src/Util/HtmlTransformer.js:142:42) at Object.<anonymous> (file:///home/ats/vc/personal/web/node_modules/.pnpm/github.com+11ty+eleventy@c39e8961a2ab7e311696d415fc4238e8fe4d4d80/node_modules/@11ty/eleventy/src/defaultConfig.js:125:13) at Object.fn (file:///home/ats/vc/personal/web/node_modules/.pnpm/github.com+11ty+eleventy@c39e8961a2ab7e311696d415fc4238e8fe4d4d80/node_modules/@11ty/eleventy/src/Benchmark/BenchmarkGroup.js:37:23) at TransformsUtil.runAll (file:///home/ats/vc/personal/web/node_modules/.pnpm/github.com+11ty+eleventy@c39e8961a2ab7e311696d415fc4238e8fe4d4d80/node_modules/@11ty/eleventy/src/Util/TransformsUtil.js:36:30)

The main problem seemed to be writing ./public/css/base.css.map. When getPlugins was called for that it was finding the map method for Array which it could not iterate over.

This fixes a couple of minor issues I saw while looking over the README file.

My main layout is written in pug, and uses an include file. When running eleventy --serve I noticed that changes to the included file would trigger rebuilds, but not show the changes. This seems to be similar to 11ty/eleventy-plugin-template-languages#3.

Here I fix this by using the list of dependencies that the pug compile function includes on the function that it returns. With this I am seeing changes to my included file reflected immediately.

When src/extension.ts is compiled the resulting file is created directly in the out/ directory, not out/src/.

When I tried to use the extension directly from the marketplace I got the following error in the developer console, and it completely broke the Enter key:

workbench.desktop.main.js:1978 Activating extension 'kaiwood.endwise' failed: Cannot find module '~/.vscode-oss/extensions/kaiwood.endwise-1.5.1-universal/out/src/extension'
Require stack:
- /usr/share/codium/resources/app/out/vs/loader.js
- /usr/share/codium/resources/app/out/bootstrap-amd.js
- /usr/share/codium/resources/app/out/bootstrap-fork.js
- .

Running from a pristine git clone resulted in a similar error, but after making this change the extension works correctly.

This may be due to me using the codium here rather than the "official" VSCode. On another computer where I do use the official release, the version from the marketplace works, but running from a git clone seems to require this change as well.

While the syntax highlighting had support for this, indentation did not. This left it indenting things way too far.

The project created by this initializer will require at least version 14 of node, so specify that in its engines declaration.

The system that I was trying this on still had node 12 as the default version, and when I attempted to run the created project it encountered import errors.

Force git's diff.mnemonicprefix option to off so that file names will be correctly parsed from the diff output.

Adjust style of descriptions, add completion of new options, and add description of the subcommand.

A couple of small changes to have branch names completed for the :Twiggy command.

Vim allows commands to be run with any unambiguous prefix, so :Twig can be used to invoke this without special handling. In fact, in common setups, if this shortcut is removed even just :Tw can be used.

The reverse operation needs to be specified on the function which has a closure around the type_shorthands in order for the reverse operation to be found by the migration builder.

• Allow using an object with schema and table name as value for the references key to setup a foreign key constraint.
• If an object with schema and table name is passed to createIndex don't include the schema when generating a name for the index.
• When generating a constraint name for a composite primary key get the table name out of an object if necessary to avoid including [object Object] in the name.

If this isn't done, errors occur when the command window is closed if delimiters were entered into the window.

Allow per-buffer configuration of base used for diffs and add GitGutterBase command to set the base and update the display.

I open up files from pull requests within vim to more closely examine the changes, and in those cases I want to use the PR target as the base for diffs to help with jumping between hunks. But, when I'm writing code the default of diffing against the index (or possibly HEAD) is more useful. Having a single setting for the base across the entire vim session can make this more difficult, also having a command to change the setting makes it easier to use tab completion.

When Vim's modeline support is enabled, the string ex: near the beginning or end of a file will be interpreted as an attempt to set options and cause errors to be displayed when the file is opened. Avoid use of this string to help out Vim users.

If the options passed to createType isn't an array treat it as a set of attributes for a composite type. This is the only form of type creation that conforms to the SQL standard.

This does leave creation of range and base types unsupported, and may make it more difficult to add support for those in the future. But, those types would generally also require defining functions which is also not supported by a helper; so anyone wanting to create those types is likely to need to use raw SQL anyway.

This seems to have been mistakenly changed in 66dbfc65.

Closes #116.

With OpenSSL 1.1.0e (likely earlier as well) the list-* commands are no longer supported causing completion to fail for subcommands and arguments of digest and cipher subcommands.

Instead use the output of openssl help, removing the section headers and splitting the list of commands which that displays. Earlier versions don't actually have a help subcommand but they did already produce the same help output for unknown subcommands, just preceded with an error message about the invalid command. This error message is removed along with the section headers.

One use of the obsolete list-cipher-algorithms subcommand remains, but it is in the completion for the genpkey subcommand which is also obsolete.

• Use <label> tags so that the apparent labels will be associated with the corresponding input fields.
• Switch to using CSS to layout the content, rather than using a table for non-tabular data.
• Add type attribute for some of the inputs, to try to help mobile browsers display the appropriate keyboard.
• Add required attribute to appropriate fields so that modern browsers will show errors if user tries to submit without filling them in.
• Use CSS to indicate optional fields to avoid repetition.

Could still use some hints for mobile browsers to display the form at a better size.

In neovim reltime() is based on the system uptime rather than the epoch so it can't be used to get the local time. Use the same fallback as is used for non-Unix systems instead. This means that %k won't support milliseconds as added in d03fc1a, but at least larger units of time should be correct.

Check list for doing migration:

• [x] Shutdown services on old pilot
  • [x] apache
  • [x] postfix
  • [x] dovecot
  • [x] mysql
  • [x] cron
  • [x] tribot
• [x] Dump mysql databases
• [x] Do a final merge of /etc repos
• [x] Add AllowUsers to /etc/ssh/sshd_config
• [x] Kick off users
• [x] Clear out or transfer postfix queue contents
• [x] Rsync content
• [x] Merge this and run the migrate playbook
• [x] Merge migrate branch of /etc repo on pilot VM (available on old pilot as well)
• [x] Restore mysql databases

Avoid including the diff in the commit message when user has commit.verbose set to yes by forcing it to no when invoking git to generate the commit message file.

Remove the port specification from one of the example table entries in the README file and add a comment explaining that the port number from the listener will be used.

I'm looking to set this up to proxy both http and https traffic using the same mappings from host name to backend, so I was curious if I'd need to duplicate that information. Reading the source of the man page I learned that this feature would let me avoid the duplication, but future people looking at this may find it nice to be able to learn that from the README file.

The survey is no longer accepting input, so it doesn't make sense ask people to provide it.

This switches the source of HTML form to an actual .html file rather than using a python module with several large strings. This allows syntax highlighting to be used when editing the source.

The success message and error messages have also been moved into the HTML source (other than in one case where it inserts the textual representation of an exception). If the registration appeared to succeed, the <form> element is simply emptied so that it will not be displayed.

Besides cleaning up the code, there is already a usability improvement. Now if an error is found the values entered into the form are preserved rather than forcing the user to go back to the previous page.

Next step will be to bring HTML into this century. Currently it's only slightly changed from what was in the .py file.

This is another character that needs to be specially handled.

Store the tab page number along with the buffer number so that windows can be swapped between different tab pages.

Closes #5.

Modify commit message parsing to respect the comment character which git has been configured to use.

If the ~/.gitconfig sets the diff.renames option to a true value 4 of the 5 tests run with rspec -d renam fail. Fix this by adding R to the list of change types included in the --diff-filter option. All tests continue to pass if this option is left at the default value.

The normal error messages aren't very helpful in determining the source of problems, so I tried using the --debug option. At first that just failed because of the issue addressed by the first commit in this series. After I addressed that it still didn't provide any help because the errors were being suppressed before the exception handler from Appl that honors that flag, so I added code to honor that flag in the main exception handlers in hermesmail.

If RbConfig has an entry for rubyarchhdrdir use it rather than joining the platform-independent rubyhdrdir with the arch. This is needed for the vendor ruby on current Debian and Ubuntu systems where the architecture dependent include files are in /usr/include/x86_64-linux-gnu/ruby-2.1.0. The old method would result in the last two path components being switched.

Fall back to using the previous method if RbConfig doesn't define rubyarchhdrdir.

Detect the comment character being used by git by looking at the first character included in the status buffer. Use that character rather than a hard-coded # when operating in :Gstatus buffers.

I don't think this is actually ready to merge yet. It fails rather spectacularly if the comment character is set to \, ^, or ]. But, as long as I don't pick one of those characters it is working well for me, along with the changes from tpope/vim-git#31. If I can get these issues fixed, I believe this would resolve #547.

Make parsing of database.yml closer to how Rails does it by preprocessing it with ERB.

Add /doc/ to .gitignore file so that using rdoc to generate docs doesn't cause git to list untracked files.

If argument being completed starts with a ^ or anything ending with two or three dots, prepend that prefix to all found ref names. This allows completing things like ^master or origin/master..master.

Add tab completion support of ref names and long options for Gitv command.

Closes #38.

Modify README to suggest use of if-shell (shortcut if) rather than run-shell (shortcut run), this makes the configuration a bit easier to read since it doesn't require as many levels of quoting and has direct support for an "else" branch rather than requiring chained boolean operators.

Also change from using tmux display-message -p to echo when retrieving the current command. The expansion is done by run or if anyway, and echo should be lighter weight than running another tmux process and is definitely shorter.

If changes are made with a cursor inside of a closed fold, the entire fold will be treated as if it was selected rather than just the matched portion of it. To avoid this open all folds around new cursors.

If changes are made with a cursor inside of a closed fold, the entire fold will be treated as if it was selected rather than just the matched portion of it. To avoid this open all folds around new cursors.

Map the "p" key to make it easy to jump to a parent of the current commit, even if there are many intervening commits on other branches, loading enough of the history to include the target commit. Allow a count to be used to specify a parent other than the first for merge commits.

Use a .gitattributes file to tell git to not highlight trailing whitespace as an error in .rst files when showing diffs. A trailing space is used on many lines in those files to indicate that the paragraph continues onto the next line.

If the g:powerline_no_python_error variable is set don't issue an error message even if vim doesn't have the required Python support. This allows a common set of configuration files to be used across systems where some of the copies of vim don't meet the requirements.

Move code to parse a .git file from the do_status method to a new method shared with get_branch_name so that branch name can be retreived from repositories which use a .git file.

Closes #634.

The "class" attribute is commonly set for <div>s, but the current snippet definition makes it difficult to create any attributes other than "id". Setup that attribute as the id is done, so that either or both may be easily included.

This series add a button to easily jump to the last frame seen on a browser using a cookie to store that information between sessions.

There are currently a couple of major issues with handling of the exclude_modes and include_modes lists for segments that prevent the latter list from working at all.

First, the results of checking the two lists are or'd together meaning that if a mode isn't in the exclude list the include list is never even checked.

Second, the check of the include list looks for the segment object in the list rather than looking for the mode identifier. That check could never succeed.

This could use a looking over before merging.

May also want to put this level in a different location, I think it would fit a bit better as rampup4 moving that existing level to rampup5, but I suspect that might cause a small problem for people who already have a cookie indicating that they've completed that level.

The term "second parent" has a specific meaning in git that is distinct from a grandparent, so avoid referring to a grandparent (~2) as a second parent (^2).

The statement that the working directory (aside: "working tree" would be a more git-like term) will always match the current state of HEAD is completely incorrect. That would imply that changes to files get in some way committed as files are saved.

Even the second part of the paragraph is incorrect. Changing HEAD does not necessarily change the contents of the working tree as in git reset without the --hard option.

The real git merge command does not use an optional second argument to specify the branch which will be modified, HEAD always determines that. The implementation here should follow that to avoid giving people mistaken ideas about how to do merges.

The first commit here modifies commands in the demonstration portion of intro/3 to remove second arguments from git merge commands. The second commit modifies the implementation of the git merge command so that it acts more like the real thing.

These changes implement full support for the ~ and ^ modifiers to references, including optional trailing numbers.

I mainly wanted to add support for ~. I'm more accustomed to using that than ^, since with my shell and configuration ^ needs to be quoted but ~ doesn't. While I was at it I decided to try adding the full support.

Here are a number of changes that I needed to make to allow me to do a build, arranged in order of my confidence from highest to lowest.

The first commit seems to be fairly clear, just requiring a change in case of the requested file name. And I have a pretty good idea of why that didn't work for me but apparently did for other people.

The second commit also appears to be a legitimate issue, but I have no idea why I seem to be the only person hit by it.

The final commit was basically a stab in the dark, but it does allow be to do a successful build.

Create a file to provide code formatting hints to editors with support for the EditorConfig format.

After updating the text send a textInput event appending a newline to it. This will cause some sites (at least StackOverflow.com) to update their preview of the rendered content. Unfortunately this also causes an extra newline to be added to the end of the area, but there tends to be an extra one there anyway.

If the x-url header is present in the request, include the %-encoded contents of that in the name of the file that is created for editing. This will allow the editor to be configured to use a different mode depending on what is being edited.

For example, I have added a file with the following to ~/.vim/ftdetect to set the filetype to markdown for editing content from stackoverflow.

" Not using setf command, since the filetype will already be set to text
" Need to override that
au BufNewFile,BufRead chrome_%%*stackoverflow*.txt  setlocal filetype=mkd

Map <C-n> and <C-p> to JumpToCommit function, which will skip over lines in the history that don't list a commit.

This function was added in commit 392cd6f with a fold marker at the end but not at the start, messing up the folding.

Here are a couple of commits ontop of pydave@f71753e for issue tpope/#7 to improve autoloading support. These changes prevent user-defined formats from triggering autoloading without requiring changes to how custom formats are defined.

Add a buffer option which, if set, will cause trailing whitespace to be stripped from lines before jumping to the next tab stop.

When used with a snippet like the following this will allow the text from the $1 tabstop to be removed and have the space in front of that automatically deleted when the user jumps to $3. This will work even if using the same key for snippet expansion and jumping to the next tabstop, whereas if the space was manually removed the key would cause the snippet to be expanded again.

snippet do
do ${1:|${2:args}|}
    $3
end
$0
endsnippet

The rvm --patch option takes an argument that is one or more file names separated by commas.

Some updates to the completion for rvm, the help output for that has changed breaking completion of subcommands. Also fix several bugs in the completion.

Having a vim modeline or emacs local variable line somewhere in the middle of the file isn't really helpful. By default, vim will only check the first and last 5 lines of a file for a modeline (assuming the modeline option is enabled). Emacs is even more strict about the type of local variable line that was in use, it will only check the first line of the file or the second line if the first line specifies a script interpreter (which isn't the case here).

Move the vim modeline to the end of the file so that it can actually be found by vim but is out of the way for editing. For emacs more work is required, convert that to the more verbose Local Variables syntax which emacs will look for starting 3000 characters from the end of the file.

Also there is no zsh mode for emacs (according to zsh-users/zsh-completions#75), use the "Shell-Script" mode instead. This seems to automatically detect that the files are for zsh.

I'm not an emacs user, so I haven't tested that portion much. But, this does at least improve the syntax highlighting there.

Here are some improvements to the completion script for ack.

If the matchit (http://www.vim.org/scripts/script.php?script_id=39) plugin is loaded, define pairs to be handled by the % key.

Change in b319569 made this comment obsolete.

Modify new_scratch_buffer method to leave the created buffer in command rather than insert mode. Currently the only use of this method is for showing stack traces for errors, when that happens it's much more likely that the user will want to use command mode to move around than to use insert mode to add text to the beginning of the buffer.

I was looking into trying to write some tests for the fold expansion problem for which I'd previously sent a change, but had the problem that even with none of my changes applied most of the tests were failing for me. Looking into this revealed that the version of screen installed on my Debian unstable system has changed how the stuff command works such that some characters are special.

To fix this problem, additional escaping needs to be done for the new version of screen. In order to do this cleanly, I did some major refactoring of how the test script interacts with vim. With this series a family of classes is used for this interaction. This has the nice additional benefit of putting all of the Windows code into a separate class. Unfortunately, the version of vim I have installed on my Windows system won't load the UltiSnips plugin so I wasn't able to completely test these changes there; but the interactions between the test script and vim do seem to be working with these changes.

Here are a couple of commits to make things less annoying when used with a copy of vim that doesn't have python support.

I like to use the same config files across a wide range of systems, and can't always guarantee that the system's vim will be compiled with python support. Any system where the vim doesn't have that support is likely to be one where I'm only doing basic things with vim, so I won't really care about snippet support. But, being constantly warned that snippets won't work gets to be annoying quickly.

Commit 13331a8 added code to show the current line after insertions, but later changes caused that to be called too soon. This left folds created by a snippet closed, and the cursor at the end of the insertion.

Cool plugin

This series add support for specifying arguments when binding a queue to an exchange. If the exchange is of type "headers" this allows routing based on message headers.

Needed on FreeBSD to get definition of uint64_t and related types.

The FreeBSD ports collection already had a similar patch.

Started looking into this while checking on the timeout test, but found that the module wouldn't compile there. I've tested that it still compiles correctly on Linux and OS X.

This test will fail without additional setup which people just installing from CPAN are unlikely to have done. So, only run this when doing tests from a git repository or an environment variable is set indicating that maintainer tests should be used.

Also add a comment indicating one possible way to let these tests pass.

If an attempt is made to undo a migration with

def change add_foreign_key :tracks, :cds end

it will currently fail with the message:

wrong number of arguments (1 for 2)

This is easily fixed by modifying the invert_* methods in the CommandRecorder module to expand the argument list that is returned in the arrays.

I've tested this for both add_foreign_key and remove_foreign_key

This branch adds trigger support to the Postgres parser and producer. Tests of the parsing have been added and all previously existing tests pass, after adjusting ones that expected that triggers would not be handled for Postgres.

I've added a new scope attribute to the Schema::Trigger class so that it can support triggers for both individual rows and complete statements.

Get this warning from Postgres producer if you try to create an index without any fields

Had some lingering text copied from doc for options method.