Skip to content

Latest commit

 

History

History
1177 lines (924 loc) · 51.7 KB

ghub.org

File metadata and controls

1177 lines (924 loc) · 51.7 KB

Ghub User and Developer Manual

Ghub provides basic support for using the APIs of various Git forges from Emacs packages. Originally it only supported the Github REST API, but now it also supports the Github GraphQL API as well as the REST APIs of Gitlab, Gitea, Gogs and Bitbucket.

Ghub abstracts access to API resources using only a handful of basic functions such as `ghub-get`. These are convenience wrappers around `ghub-request`. Additional forge-specific wrappers like `glab-put`, `gtea-put`, `gogs-post` and `buck-delete` are also available. Ghub does not provide any resource-specific functions, with the exception of `FORGE-repository-id`.

When accessing Github, then Ghub handles the creation and storage of access tokens using a setup wizard to make it easier for users to get started. The tokens for other forges have to be created manually.

Ghub is intentionally limited to only provide these two essential features — basic request functions and guided setup — to avoid being too opinionated, which would hinder wide adoption. It is assumed that wide adoption would make life easier for users and maintainers alike, because then all packages that talk to forge APIs could be configured the same way.

This manual is for Ghub version 3.3.0 (v3.3.0-1-g4ebe60b+1).

Copyright (C) 2017-2020 Jonas Bernoulli <jonas@bernoul.li>

You can redistribute this document and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This document is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

Introduction

Ghub provides basic support for using the APIs of various Git forges from Emacs packages. Originally it only supported the Github REST API, but now it also supports the Github GraphQL API as well as the REST APIs of Gitlab, Gitea, Gogs and Bitbucket.

Ghub abstracts access to API resources using only a handful of basic functions such as `ghub-get`. These are convenience wrappers around `ghub-request`. Additional forge-specific wrappers like `glab-put`, `gtea-put`, `gogs-post` and `buck-delete` are also available. Ghub does not provide any resource-specific functions, with the exception of `FORGE-repository-id`.

When accessing Github, then Ghub handles the creation and storage of access tokens using a setup wizard to make it easier for users to get started. The tokens for other forges have to be created manually.

Ghub is intentionally limited to only provide these two essential features — basic request functions and guided setup — to avoid being too opinionated, which would hinder wide adoption. It is assumed that wide adoption would make life easier for users and maintainers alike, because then all packages that talk to forge APIs could be configured the same way.

Fancier interfaces can be implemented on top of Ghub, and one such wrapper — named simply Ghub+ — has already been implemented. The benefit of basing various opinionated interfaces on top of a single library that provides only the core functionality is that choosing the programming interface no longer dictates how access tokens are handled. Users can then use multiple packages that access the Github API without having to learn the various incompatible ways packages expect the appropriate token to be made available to them.

Ghub uses the built-in auth-source library to store access tokens. That library is very flexible and supports multiple backends, which means that it is up to the user how secrets are stored. They can, among other things, choose between storing secrets in plain text for ease of use, or encrypted for better security.

Previously (as in until this library is widely adopted) it was up to package authors to decide if things should be easy or secure. (Note that auth-source defaults to “easy” — you have been warned.)

Ghub expects package authors to use a dedicated access token instead of sharing a single token between all packages that rely on it. That means that users cannot configure Ghub once and later start using a new package without any additional setup. But Ghub helps with that.

When the user invokes some command that ultimately results in ghub-request being called and the appropriate token is not available yet, then the user is guided through the process of creating and storing a new token, and at the end of that process the request is carried out as if the token had been available to begin with.

Getting Started

_

Each package that uses Ghub uses its own token. Despite that, chances are good that after successfully configuring one package you can just start using another package pretty much instantly.

If the necessary token to access a Github instance is not available when a package makes an API request, then a setup wizard pops up, and after answering a few questions you are good to go. Even the request that caused the wizard to be summoned should succeed and for most users this should be true even when configuring the very first token.

However, in some situations some manual configuration is necessary before using the wizard, or the wizard cannot be used at all:

  • If you don’t want to use the wizard then you don’t have to and can create tokens manually as described in <a href=”*Manually Creating and Storing a Token”>*Manually Creating and Storing a Token.
  • Unfortunately only Github supports the creation of tokens by using the API. If you want to access another forge, then you have to create the token manually as described in <a href=”*Manually Creating and Storing a Token”>*Manually Creating and Storing a Token. Also see *Support for Other Forges.
  • If you want to access a Github Enterprise instance, then you have to tell Ghub about that before the wizard makes its appearance by setting the Git variable github.host. You also have to tell Ghub your username for that instance using the variable github.HOST.user even if it is the same as on Github.com.

    These variables are documented in *Configuration Variables. Also see *Setting the Username. TL;DR: If your Github Enterprise instance is hosted at git.example.com and your username on that instance is jtribbiani, then set github.host to git.example.com/api/v3 in every repository cloned from that instance (i.e. do not set it globally) and globally set github.git.example.com/api/v3.user to jtribbiani. The latter is necessary even if your username on Github.com is the same.

    For some Github Enterprise instances /api has to be used instead of /api/v3 as shown above. If the wizard fails in both cases, then try creating the token manually.

  • If the variable github.user (or github.HOST.user for an Enterprise instance) is unset when the wizard is first summoned, then you are asked to provide your username. That value is then stored globally to avoid having to ask you that question once per repository. If you have multiple accounts on Github.com (or a Github Enterprise instance), then you have to explicitly tell Ghub about that. This can be done by setting the repository-local values of the appropriate variable before the wizard is invoked.
  • You might forget to do the above, which is why it is important to carefully read the output of the wizard. If it turns out that you forgot to set a variable, then you must abort, set the variable, and repeat the request to trigger the wizard again.
  • The setup wizard should work even if you have enabled two-factor authentication. However, if your Github Enterprise instance enforces Single Sign-On as an additional security measure, then you are out of luck and have to create the token manually as described in *Manually Creating and Storing a Token.

The variables mentioned above — and others — are documented in *Configuration Variables and the setup wizard is documented in *Interactively Creating and Storing a Token.

Setting the Username

If you haven’t set the Git variable github.user yet when making a request, then you will be asked:

Git variable `github.user' is unset.  Set to:

You are expected to provide your Github username here. The provided value will be saved globally (using git config --global github.user USERNAME).

If you need to identify as another user in a particular repository, then you have to set that variable locally, before making a request:

cd /path/to/repo
git config github.user USERNAME

For Github Enterprise instances you have to specify where the API can be accessed before you try to access it and a different variable has to be used to set the username. For example if the API is available at https://example.com/api/v3, then you should do this:

# Do this once
git config --global github.example.com/api/v3.user EMPLOYEE

# Do this for every corporate repository
cd /path/to/repo
git config github.host example.com/api/v3

If you do not set github.example.com/api/v3.user, then you will be asked to provide the value when trying to make a request, but you do have to manually set github.host, or Ghub assumes that you are trying to access api.github.com.

Interactively Creating and Storing a Token

Ghub uses a different token for every package as well as for every machine from which you access the Github API (and obviously also for every Github instance and user). This allows packages to only request the scopes that they actually need and also gives users the opportunity to refuse access to certain scopes if they expect to not use the features that need them.

Usually you don’t have to worry about creating and storing a token yourself and can just make a request. Note however that you don’t have to use the setup wizard described below. Alternatively you can perform the setup manually as described in the next section.

If you make a request and the required token is not available yet, then the setup wizard will first ask you something like this:

Such a Github API token is not available:

  Host:    api.github.com
  User:    USERNAME
  Package: PACKAGE

  Scopes requested in `PACKAGE-github-token-scopes':
    repo
  Store on Github as:
    "Emacs package PACKAGE @ LOCAL-MACHINE"
  Store locally according to option `auth-sources':
    ("~/.authinfo" "~/.authinfo.gpg" "~/.netrc")

If in doubt, then abort and first view the section of
the Ghub documentation called "Interactively Creating
and Storing a Token".

Create and store such a token? (yes or no)

If you don’t have any doubts, then answer “yes”. Lets address some of the doubts that you might have:

  • Host usually is “api.github.com” and that is usually what you want. If you are trying to access a Github Enterprise instance, then it should be something else and you have to set the value manually before the setup wizard is summoned, as described in the parent section.
  • User should be your Github.com (or Github Enterprise instance) username. If it is something else and it doesn’t look like a simple typo, then you should read the parent section again. In either case you have to abort.
  • Package should be the name of the package you are using to access the Github API.

    If it is ghub, then the package author disregarded that convention and you should probably report a bug in the issue tracker of that package.

    Or you yourself are using ghub-request or one of its wrappers directly, in which case this is expected and perfectly fine. In that case you might however want to abort and change the value of the variable ghub-github-token-scopes before triggering the wizard again.

  • Each PACKAGE has to specify the tokens that it needs using a variable named PACKAGE-github-token-scopes. The doc-string of that variable should document why the various scopes are needed.

    The meaning of the various scopes are documented at https://magit.vc/goto/f63aeb0a.

  • The value of auth-sources is shown. The default value causes secrets to be stored in plain text. Because this might be unexpected, Ghub additionally displays a warning when appropriate. 
    WARNING: The token will be stored unencrypted in "~/.authinfo".
         If you don't want that, you have to abort and customize
         the `auth-sources' option.

    Whether that is something that needs fixing, is up to you. If your answer is yes, then you should abort and see <a href=”*How Ghub uses Auth-Source”>*How Ghub uses Auth-Source for instructions on how to save the token more securely.

  • When creating a token, it is necessary to provide a token description. Ghub uses descriptions that have the form “Emacs package PACKAGE @ LOCAL-MACHINE”.

    Github uses the token description to identify the token, not merely as something useful to humans. Token descriptions therefore have to be unique and in rare cases you get an additional prompt, asking you something like:

    A token named "Emacs package PACKAGE @ LOCAL-MACHINE"
already exists on Github.  Replace it?

    You might see this message when you have lost the old token and want to replace it with a new one, in which case you should obviously just proceed.

    Or two of your computers have the same hostname, which is bad practice because it gains you nothing but leads to issues such as this. Or you are dual-booting on this machine and use the same hostname in all operating systems, which is a somewhat reasonable thing to do, but never-the-less leads to issues like this.

    In either case you will have to use something other than the value returned by system-name to identify the current machine or operating system. Or you can continue to identify different things using the same identifier, in which case you have to manually distribute the token.

    The former is recommended and also easier to do, using the variable ghub-override-system-name. See *Configuration Variables for details.

After the above prompt you are also asked for your username and password. If you have enabled two-factor authentication, then you also have to provide the authentication code at least twice. If you make sure the code is still good for a while when asked for it first, then you can just press RET at the later prompt(s).

Manually Creating and Storing a Token

If you cannot or don’t want to use the wizard then you have to (1) figure out what scopes a package wants, (2) create such a token using the web interface and (3) store the token where Ghub expects to find it.

A package named PACKAGE has to specify the scopes that it wants in the variable named PACKAGE-github-token-scopes. The doc-string of such variables should document what the various scopes are needed for.

To create or edit a token go to https://github.com/settings/tokens. For Gitlab.com use https://gitlab.com/profile/personal_access_tokens.

Finally store the token in a place where Ghub looks for it, as described in *How Ghub uses Auth-Source.

If you store the token in a file like ~~/.authinfo~, then note that auth-source’s parsing of that file is brittle. Make sure the file ends with a newline character, that there are no empty or invalid lines, and that all comments are prefixed with #.

How Ghub uses Auth-Source

Please see info:auth for all the gory details about Auth-Source. Some Ghub-specific information and important notes follow.

The variable auth-sources controls how and where Auth-Source stores new secrets and where it looks for known secrets. The default value is ("~/.authinfo" "~/.authinfo.gpg" "~/.netrc"), which means that it looks in all of these files in order to find secrets and that it stores new secrets in ~~/.authinfo~ because that is the first element of the list. It doesn’t matter which files already do or don’t exist when storing a new secret, the first file is always used.

Secrets are stored in ~~/.authinfo~ in plain text. If you don’t want that (good choice), then you have to customize auth-sources, e.g. by flipping the positions of the first two elements.

Auth-Source also supports storing secrets in various key-chains. Refer to its documentation for more information.

Some Auth-Source backends only support storing three values per entry, the “machine”, the “login” and the “password”. Because Ghub uses separate tokens for each package, it has to squeeze four values into those three slots, and it does that by using “USERNAME^PACKAGE” as the “login”.

Assuming your username is “ziggy”,the package is named “stardust”, and you want to access Github.com an entry in one of the three mentioned files would then look like this:

machine api.github.com login ziggy^stardust password 012345abcdef...

Assuming your username is “ziggy”,the package is named “stardust”, and you want to access Gitlab.com an entry in one of the three mentioned files would then look like this:

machine gitlab.com/api/v4 login ziggy^stardust password 012345abcdef...

Using Ghub in Personal Scripts

You can use ghub-request and its wrapper functions in your personal scripts, of course. Unlike when you use Ghub from a package that you distribute for others to use, you don’t have to specify a package in personal scripts.

;; This is perfectly acceptable in personal scripts ...
(ghub-get "/user")

;; ... and actually equal to
(ghub-get "/user" nil :auth 'ghub)

;; In packages you have to specify the package using AUTH.
(ghub-get "/user" nil :auth 'foobar)

When you do not specify the AUTH argument, then a request is made on behalf of the ghub package itself. Like for any package that uses Ghub, ghub has to declare what scopes it needs, using, in this case, the variable ghub-github-token-scopes.

The default value of that variable is (repo) and you might want to add additional scopes. You can later add additional scopes to an existing token, using the web interface at https://github.com/settings/tokens.

If you do that, then you might want to also set the variable accordingly, but note that Ghub only consults that when creating a new token. If you want to know a token’s effective scopes use the command ghub-token-scopes, described in the next section.

Using Ghub in a Package

Every package should use its own token. This allows you as the author of some package to only request access to API scopes that are actually needed, which in turn might make it easier for users to trust your package not to do unwanted things.

The scopes used by PACKAGE have to be defined using the variable PACKAGE-github-token-scopes, and you have to tell ghub-request on behalf of which package a request is being made by passing the symbol PACKAGE as the value of its AUTH argument.

(ghub-request "GET" "/user" nil :auth 'PACKAGE)
  • Variable: PACKAGE-github-token-scopes

    This variable defines the token scopes requested by the package named PACKAGE. The doc-string should explain what the various scopes are needed for to prevent users from giving PACKAGE fewer permissions than it absolutely needs and also to give them greater confidence that PACKAGE is only requesting the permissions that it actually needs.

    The value of this variable does not necessarily correspond to the scopes that the respective token actually gives access to. There is nothing that prevents users from changing the value after creating the token or from editing the token’s scopes later on.

    So it is pointless to check the value of this variable before making a request. You also should not query the API to reliably determine the supported tokens before making a query. Doing the latter would mean that every request becomes two requests and that the first request would have to be done using the user’s password instead of a token.

  • Command: ghub-token-scopes

    Because we cannot be certain that the user hasn’t messed up the scopes, Ghub provides this command to make it easy to debug such issues without having to rely on users being thoughtful enough to correctly determine the used scopes manually.

    Just tell users to run M-x ghub-token-scopes and to provide the correct values for the HOST, USERNAME and PACKAGE when prompted, and to then post the output.

    It is to be expected that users will occasionally mess that up so this command outputs not only the scopes but also the user input so that you can have greater confidence in the validity of the user’s answer.

    Scopes for USERNAME^PACKAGE@HOST: (SCOPE...)

API

_

This section describes the Ghub API. In other words it describes the public functions and variables provided by the Ghub package and not the APIs of the supported forges, which can be accessed by using those functions. The forge APIs are documented at:

Making Requests

  • Function: ghub-request method resource &optional params &key query payload headers unpaginate noerror reader username auth host callback errorback url value error extra method*

    This function makes a request for RESOURCE using METHOD. PARAMS, QUERY, PAYLOAD and/or HEADERS are alists holding additional request data. The response body is returned and the response headers are stored in the variable ghub-response-headers.

    • METHOD is the HTTP method, given as a string.
    • RESOURCE is the resource to access, given as a string beginning with a slash.
    • PARAMS, QUERY, PAYLOAD and HEADERS are alists and are used to specify request data. All these arguments are alists that resemble the JSON expected and returned by the Github API. The keys are symbols and the values stored in the cdr (not the cadr) can be strings, integers, or lists of strings and integers.

      The Github API documentation is vague on how data has to be transmitted and for a particular resource usually just talks about “parameters”. Generally speaking when the METHOD is “HEAD” or “GET”, then they have to be transmitted as a query, otherwise as a payload.

      • Use PARAMS to automatically transmit like QUERY or PAYLOAD would depending on METHOD.
      • Use QUERY to explicitly transmit data as a query.
      • Use PAYLOAD to explicitly transmit data as a payload. Instead of an alist, PAYLOAD may also be a string, in which case it gets encoded as UTF-8 but is otherwise transmitted as-is.
      • Use HEADERS for those rare resources that require that the data is transmitted as headers instead of as a query or payload. When that is the case, then the Github API documentation usually mentions it explicitly.
    • If SILENT is non-nil, then progress reports and the like are not messaged.
    • If UNPAGINATE is t, then this function makes as many requests as necessary to get all values. If UNPAGINATE is a natural number, then it gets at most that many pages. For any other non-nil value it raises an error.
    • If NOERROR is non-nil, then no error is raised if the request fails and nil is returned instead. If NOERROR is return, then the error payload is returned instead of nil.
    • If READER is non-nil, then it is used to read and return from the response buffer. The default is ghub--read-json-payload. For the very few resources that do not return JSON, you might want to use ghub--decode-payload.
    • If USERNAME is non-nil, then the request is made on behalf of that user. It is better to specify the user using the Git variable github.user for “api.github.com”, or github.HOST.user if connecting to a Github Enterprise instance.
    • Each package that uses Ghub should use its own token. If AUTH is nil or unspecified, then the generic ghub token is used instead. This is only acceptable for personal utilities. A package that is distributed to other users should always use this argument to identify itself, using a symbol matching its name.

      Package authors who find this inconvenient should write a wrapper around this function and possibly for the method-specific functions as well.

      Beside nil, some other symbols have a special meaning too. none means to make an unauthorized request. basic means to make a password based request. If the value is a string, then it is assumed to be a valid token. basic and an explicit token string are only intended for internal and debugging uses.

      If AUTH is a package symbol, then the scopes are specified using the variable AUTH-github-token-scopes. It is an error if that is not specified. See ghub-github-token-scopes for an example.

    • If HOST is non-nil, then connect to that Github instance. This defaults to “api.github.com”. When a repository is connected to a Github Enterprise instance, then it is better to specify that using the Git variable github.host instead of using this argument.
    • If FORGE is gitlab, then connect to Gitlab.com or, depending on HOST, to another Gitlab instance. This is only intended for internal use. Instead of using this argument you should use function glab-request and other glab-* functions.
    • If CALLBACK and/or ERRORBACK is non-nil, then this function makes one or more asynchronous requests and calls CALLBACK or ERRORBACK when finished. If no error occurred, then it calls CALLBACK, unless that is nil.

      If an error occurred, then it calls ERRORBACK, or if that is nil, then CALLBACK. ERRORBACK can also be t, in which case it signals instead. NOERROR is ignored for all asynchronous requests.

      Both callbacks are called with four arguments.

      1. For CALLBACK, the combined value of the retrieved pages. For ERRORBACK, the error that occurred when retrieving the last page.
      2. The headers of the last page as an alist.
      3. Status information provided by url-retrieve. Its :error property holds the same information as the first argument to ERRORBACK.
      4. A ghub--req struct, which can be passed to ghub-continue (which see) to retrieve the next page, if any.
  • Function: ghub-continue args

    If there is a next page, then this function retrieves that.

    This function is only intended to be called from callbacks. If there is a next page, then that is retrieved and the buffer that the result will be loaded into is returned, or t if the process has already completed. If there is no next page, then return nil.

    Callbacks are called with four arguments (see ghub-request). The forth argument is a ghub--req struct, intended to be passed to this function. A callback may use the struct’s extra slot to pass additional information to the callback that will be called after the next request. Use the function ghub-req-extra to get and set the value of that slot.

    As an example, using ghub-continue in a callback like so:

    (ghub-get "/users/tarsius/repos" nil
          :callback (lambda (value _headers _status req)
                      (unless (ghub-continue req)
                        (setq my-value value))))

    is equivalent to:

    (ghub-get "/users/tarsius/repos" nil
          :unpaginate t
          :callback (lambda (value _headers _status _req)
                      (setq my-value value)))

    To demonstrate how to pass information from one callback to the next, here we record when we start fetching each page:

    (ghub-get "/users/tarsius/repos" nil
          :extra (list (current-time))
          :callback (lambda (value _headers _status req)
                      (push (current-time) (ghub-req-extra req))
                      (unless (ghub-continue req)
                        (setq my-times (ghub-req-extra req))
                        (setq my-value value))))
  • Variable: ghub-response-headers

    A select few Github API resources respond by transmitting data in the response header instead of in the response body. Because there are so few of these inconsistencies, ghub-request always returns the response body.

    To access the response headers use this variable after ghub-request has returned.

  • Function: ghub-response-link-relations req headers payload

    This function returns an alist of the link relations in HEADERS, or if optional HEADERS is nil, then those in ghub-response-headers.

    When accessing a Bitbucket instance then the link relations are in PAYLOAD instead of HEADERS, making their API merely RESTish and forcing this function to append those relations to the value of ghub-response-headers, for later use when this function is called with nil for PAYLOAD.

  • Variable: ghub-override-system-name

    If non-nil, the value of this variable is used to override the value returned by system-name for the purpose of identifying the local machine, which is necessary because Ghub uses separate tokens for each machine. Also see *Configuration Variables.

  • Variable: ghub-github-token-scopes
  • Variable: PACKAGE-github-token-scopes

    Such a variable defines the token scopes requested by the respective package PACKAGE given by the first word in the variable name. ghub itself is treated like any other package. Also see <a href=”*Using Ghub in a Package”>*Using Ghub in a Package.

  • Function: ghub-head resource &optional params &key query payload headers unpaginate noerror reader username auth host callback errorback
  • Function: ghub-get resource &optional params &key query payload headers unpaginate noerror reader username auth host callback errorback

    These functions are simple wrappers around ghub-request. Their signature is identical to that of the latter, except that they do not have an argument named METHOD. The HTTP method is instead given by the second word in the function name.

    As described in the documentation for ghub-request, it depends on the used method whether the value of the PARAMS argument is used as the query or the payload. For the “HEAD” and “GET” methods it is used as the query.

  • Function: ghub-put resource &optional params &key query payload headers unpaginate noerror reader username auth host callback errorback
  • Function: ghub-post resource &optional params &key query payload headers unpaginate noerror reader username auth host callback errorback
  • Function: ghub-patch resource &optional params &key query payload headers unpaginate noerror reader username auth host callback errorback
  • Function: ghub-delete resource &optional params &key query payload headers unpaginate noerror reader username auth host callback errorback

    These functions are simple wrappers around ghub-request. Their signature is identical to that of the latter, except that they do not have an argument named METHOD. The HTTP method is instead given by the second word in the function name.

    As described in the documentation for ghub-request, it depends on the used method whether the value of the PARAMS argument is used as the query or the payload. For the “PUT”, “POST”, “PATCH” and “DELETE” methods it is used as the payload.

  • Function: ghub-wait resource &optional duration &key username auth host

    Some API requests result in an immediate successful response even when the requested action has not actually been carried out yet. An example is the request for the creation of a new repository, which doesn’t cause the repository to immediately become available. The Github API documentation usually mentions this when describing an affected resource.

    If you want to do something with some resource right after making a request for its creation, then you might have to wait for it to actually be created. This function can be used to do so. It repeatedly tries to access the resource until it becomes available or until a timeout is reached. In the latter case it signals ghub-error.

    RESOURCE specifies the resource that this function waits for.

    DURATION specifies the maximum number of seconds to wait for, defaulting to 64 seconds. Emacs will block during that time, but the user can abort using C-g.

    The first attempt is made immediately and will often succeed. If not, then another attempt is made after two seconds, and each subsequent attempt is made after waiting as long as we already waited between all preceding attempts combined.

    See ghub-request’s documentation above for information about the other arguments.

  • Function: ghub-graphql graphql &optional variables &key username auth host callback

    This function makes a GraphQL request using GRAPHQL and VARIABLES as inputs. GRAPHQL is a GraphQL string. VARIABLES is a JSON-like alist. The other arguments behave as for ghub-request (which see).

    The response is returned as a JSON-like alist. Even if the response contains errors, this function does not raise an error. Cursor-handling is likewise left to the caller.

Authentication

  • Command: ghub-create-token

    This command creates a new token using the values it reads from the user and then stores it according to the variable auth-sources. It can also be called non-interactively, but you shouldn’t do that yourself.

    This is useful if you want to fully setup things before attempting to make the initial request, if you want to provide fewer than the requested scopes or customize auth-sources first, or if something has gone wrong when using the wizard that is used when making a request without doing this first. (Note that instead of using this command you can also just repeat the initial request after making the desired adjustments — that is easier.)

    This command reads, in order, the HOST (Github instance), the USERNAME, the PACKAGE, and the SCOPES in the minibuffer, providing reasonable default choices. SCOPES defaults to the scopes that PACKAGE requests using the variable PACKAGE-github-token-scopes.

  • Command: ghub-token-scopes

    Users are free to give a token access to fewer scopes than what the respective package requested. That can, of course, lead to issues, and package maintainers have to be able to quickly determine if such a (mis-)configuration is the root cause when users report issues.

    This command reads the required values in the minibuffer and then shows a message containing these values along with the scopes of the respective token. It also returns the scopes (only) when called non-interactively. Also see *Using Ghub in a Package.

Configuration Variables

The username and, unless you only use Github.com itself, the Github Enterprise instance have to be configured using Git variables. In rare cases it might also be necessary to specify the identity of the local machine, which is done using a lisp variable.

  • Variable: github.user

    The Github.com username. This should be set globally and if you have multiple Github.com user accounts, then you should set this locally only for those repositories that you want to access using the secondary identity.

  • Variable: github.HOST.user

    This variable serves the same purpose as github.user but for the Github Enterprise instance identified by HOST.

    The reason why separate variables are used is that this makes it possible to set both values globally instead of having to set one of the values locally in each and every repository that is connected to the Github Enterprise instance, not Github.com.

  • Variable: github.host

    This variable should only be set locally for a repository and specifies the Github Enterprise edition that that repository is connected to. You should not set this globally because then each and every repository becomes connected to the specified Github Enterprise instance, including those that should actually be connected to Github.com.

    When this is undefined, then “api.github.com” is used (defined in the constant ghub-default-host, which you should never attempt to change.)

  • Variable: ghub-override-system-name

    Ghub uses a different token for each quadruple (USERNAME PACKAGE HOST LOCAL-MACHINE). Theoretically it could reuse tokens to some extent but that would be more difficult to implement, less flexible, and less secure (though slightly more convenient).

    A token is identified on the respective Github instance (Github.com or a Github Enterprise instance) using the pair (PACKAGE . LOCAL-MACHINE), or more precisely the string “Emacs package PACKAGE @ LOCAL-MACHINE”. USERNAME and HOST do not have to be encoded because the token is stored for USERNAME on HOST and cannot be used by another user and/or on another instance.

    There is one potential problem though; for any given (PACKAGE . LOCAL-MACHINE) there can only be one token identified by “Emacs package PACKAGE @ LOCAL-MACHINE”; Github does not allow multiple tokens with the same description because it uses the description as the identifier (it could use some hash instead, but alas it does not).

    If you have multiple machines and some of them have the same name, then you should probably change that as this is not how things ought to be. However if you dual-boot, then it might make sense to give that machine the same name regardless of what operating system you have booted into.

    You could use the same token on both operating systems, but setting that up might be somewhat difficult because it is not possible to download an existing token from Github. You could, of course, locally copy the token, but that is inconvenient and would make it harder to only revoke the token used on your infected Windows installation without also revoking it for your totally safe *BSD installation.

    Alternatively you can set this variable to a unique value, that will then be used to identify the local machine instead of the value returned by system-name.

GraphQL Support

  • Function: ghub-graphql graphql &optional variables &key username auth host callback silent callback errorback value extra

    This function makes a GraphQL request using GRAPHQL and VARIABLES as inputs. GRAPHQL is a GraphQL string. VARIABLES is a JSON-like alist. The other arguments behave as for ghub-request (which see).

    The response is returned as a JSON-like alist. Even if the response contains errors, this function does not raise an error. Cursor-handling is likewise left to the caller.

ghub-graphql is a thin convenience wrapper around ghub-request, similar to ghub-post and friends. While the latter only hard-code the value of the METHOD argument, the former also hard-codes RESOURCE and constructs PAYLOAD from GRAPHEQL and VARIABLES. It also drops UNPAGINATE, NOERROR, READER (internal functions expect alist-ified JSON) and FORGE (only Github currently supports GraphQL).

ghub-graphql does not account for the fact that pagination works differently in GraphQL than it does in REST, so users of this function have to deal with that themselves. Likewise error handling works differently and has to be done by the caller too.

An early attempt at implementing automatic unpaginating for GraphQL can be found in the faithful-graphql branch, provided I haven’t deleted that by now. On that branch I try to do things as intended by the designers of GraphQL, using variables and fragments, and drowning in a sea of boilerplate.

The problem with that approach is that it only works for applications that fetch specific information on demand and actually want things to be paginated. I am convinced that GraphQL is very nice for web apps.

However the Forge package for which I am implementing all of this has very different needs. It wants to fetch “all the data” and “cache” it locally, so that it is available even when there is no internet connection. GraphQL was designed around the idea that you should be able to “ask for what you need and get exactly that”. But when that boils down to “look, if I persist, then you are going to hand me over all the data anyway, so just caught it up already”, then things start to fall apart. If Github’s GraphQL allowed pagination to be turned off completely, then teaching ghub-graphql about error handling would be enough.

But it doesn’t and when doing things as intended, then that leads to huge amounts of repetitive boilerplate, which is so boring to write that doing it without introducing bugs left and right is near impossible; so I decided to give up on GraphQL variables, fragments and conditions, and instead implement something more powerful, though also more opinionated.

  • Function: ghub–graphql-vacuum query variables callback &optional until &key narrow username auth host forge

    This function is an opinionated alternative to ghub-graphql. It relies on dark magic to get the job done.

    It makes an initial request using QUERY. It then looks for paginated edges in the returned data and makes more requests to resolve them. In order to do so it automatically transforms the initial QUERY into another query suitable for that particular edge. The data retrieved by subsequent requests is then injected into the data of the original request before that is returned or passed to the callback. If subsequently retrieved data features new paginated edges, then those are followed recursively.

    The end result is essentially the same as using ghub-graphql, if only it were possible to say “do not paginate anything”. The implementation is much more complicated because it is not possible to do that.

    QUERY is a GraphQL query expressed as an s-expression. The graphql package is used to turn that into a GraphQL query string, but the format is somewhat different than as documented for that package. Also only a subset of the GraphQL features are supported; fragments for example are not, and magical stuff happens to variables. This is not documented yet, I am afraid. Look at existing callers.

    VARIABLES is a JSON-like alist as for ghub-graphql.

    UNTIL is an alist ((EDGE-until . VALUE)...). When unpaginating EDGE try not to fetch beyond the element whose first field has the value VALUE and remove that element as well as all “lesser” elements from the retrieved data if necessary. Look at forge--pull-repository for an example. This is only useful if you “cache” the response locally and want to avoid fetching data again that you already have.

    Other arguments behave as for ghub-graphql and ghub-request, more or less.

Using ghub--graphql-vacuum, the following resource specific functions are implemented. These functions are not part of the public API yet and are very much subject to change.

  • Function: ghub-fetch-repository owner name callback &optional until &key username auth host forge

    This function asynchronously fetches forge data about the specified repository. Once all data has been collected, CALLBACK is called with the data as the only argument.

  • Function: ghub-fetch-issue owner name callback &optional until &key username auth host forge

    This function asynchronously fetches forge data about the specified issue. Once all data has been collected, CALLBACK is called with the data as the only argument.

  • Function: ghub-fetch-pullreq owner name callback &optional until &key username auth host forge

    This function asynchronously fetches forge data about the specified pull-request. Once all data has been collected, CALLBACK is called with the data as the only argument.

Note that in order to avoid duplication all of these functions base their initial query on the query stored in ghub-fetch-repository. The latter two pass that query through ghub--graphql-prepare-query, which then used ghub--graphql-narrow-query to remove parts the caller is not interested in. These two functions are also used internally, when unpaginating, but as demonstrated here they can be useful even before making an initial request.

Support for Other Forges

Forge Functions and Variables

Originally Ghub supported only Github but now it also supports Gitlab, Gitea, Gogs and Bitbucket. The function ghub-request and all the ghub-METHOD convenience wrappers default to acting on a Github forge but can be told to act on another forge using their FORGE argument.

The FORGE argument only specifies what kind of forge to act on, not which instance. The HOST argument can be used to select the instance. For some forges a default instance is defined:

  • Forge github defaults to host api.github.com.
  • Forge gitlab defaults to host gitlab.com/api/v4.
  • Forge bitbucket defaults to host api.bitbucket.org/2.0.
  • No canonical host exists for the gitea and gogs forges and localhost:3000/api/v1 is used as the default host in both cases.

Together the FORGE and HOST arguments specify the forge type and instance. In addition to that, it is also necessary to specify on whose behalf the request is being made, which can be done using the USERNAME and AUTH arguments.

Having to specify these arguments for every request is inconvenient. Additional variables and convenience functions can be used to make that unnecessary in most cases.

These variables can be set globally and/or for a specific repository as explained in *Configuration Variables with a focus on Github instances. To summarize:

  • For https://github.com the Git variable github.user specifies the user.
  • For another github instance the Git variable github.HOST.user specifies the user. The HOST in that variable name is the same as the value of the HOST argument of the called function.
  • Instead of specifying the HOST in every function call, the Git variable github.host can be used. This should only be set locally.

For gitlab and bitbucket forges similar variables are available:

  • gitlab.user specifies the https://gitlab.com user.
  • gitlab.HOST.user specifies the user for the HOST gitlab instance.
  • gitlab.host specifies the gitlab host, unless the HOST argument is non-nil
  • bitbucket.user specifies the https://bitbucket.org user.
  • bitbucket.HOST.user specifies the user for the HOST bitbucket instance.
  • bitbucket.host specifies the bitbucket host, unless the HOST argument is non-nil.

For gitea and gogs forges some similar variables are available, however for some of the ghub.* variables no equivalent variable exist for these two forges:

  • gitea.user is not used because no canonical gitea instance exists.
  • gitea.HOST.user specifies the user for the HOST gitea instance.
  • gitea.host specifies the gitea host, unless the HOST argument is non-nil
  • gogs.user is not used because no canonical gitea instance exists.
  • gogs.HOST.user specifies the user for the HOST gogs instance.
  • gogs.host specifies the gogs host, unless the HOST argument is non-nil

ghub-request and ghub-METHOD can be used to make a request for any of the supported forge types, but except when making a request for a github instance, then that requires the use of the FORGE argument.

To avoid that, functions named FORGE-request and FORGE-METHOD are also available. The following forms are equivalent, for example:

(ghub-get ... :auth 'PACKAGE :forge 'gitlab)
(glab-get ... :auth 'PACKAGE)

These forms would remain equivalent even if you did not specify a value for the AUTH arguments — but you should not do that if you plan to share your code with others (see *Using Ghub in a Package). If you do omit AUTH, then the request is made on behalf of the ghub package, regardless of the symbol prefix of the function you use to do so.

All FORGE-request and FORGE-METHOD functions, including but not limited to ghub-METHOD, are very simple wrappers around ghub-request. They take fewer arguments than ghub-request and instead pass constant values for the arguments METHOD and/or FORGE.

Forge Limitations and Notes

  • The token creation wizard is only available for github forges, because all other forges do not support using the API to create an API token. As a consequence, if the user makes a request and the necessary token cannot be found, then that results in an error. Tokens can be created at:

    Also see *Manually Creating and Storing a Token and <a href=”*How Ghub uses Auth-Source”>*How Ghub uses Auth-Source.

  • As mentioned in the previous node, the variables gitea.host and gogs.host are not taken into account.
  • Gitea and Gogs do not support limiting a token to certain scopes.
  • The Bitbucket API is fairly broken. Some resources only work if a slash is appended while others only work if no slash is appended. I am unable to access any private repositories and some resources don’t work for me at all. Also the API is only RESTish; pagination information is part of the response body instead of the header. Due to such issues it is possible that I will eventually have to remove support for Bitbucket altogether.
  • The Gitlab API documentation is not always accurate, though I don’t have an example at hand. It also isn’t structured well, making it occasionally difficult to find the information one is looking for.
  • Where one would use user/repo when accessing another forge, one has to use user%2Frepo when accessing Gitlab, e.g.: 
    (glab-get "/projects/python-mode-devs%2Fpython-mode")

_ Copying

Copyright (C) 2017-2020 Jonas Bernoulli <jonas@bernoul.li>

You can redistribute this document and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This document is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

_