diff mbox series

[PATCHv2] docs/manual: using a branch name as FOO_VERSION does not work

Message ID 20180511155058.26875-1-yann.morin.1998@free.fr
State Accepted
Headers show
Series [PATCHv2] docs/manual: using a branch name as FOO_VERSION does not work | expand

Commit Message

Yann E. MORIN May 11, 2018, 3:50 p.m. UTC
For various reasons, we've always suggested users to avoid using a
branch as version string for their packages, because it does not work
as a they would expect:

  - it is not reproducible, because the branch may change between two
    builds that are done at different times;

  - it does not even follow the branch, as Buildroot anyway generates
    a local tarball, which it will reuse on subsequent builds.
    Furthermore, since we fetch and not pull, any existing local branch
    is not updated.

Yet, until recently, using a branch name would just work (with the
above limitations): the git tree was cloned, the branch checked out,
and the tarball created.

But with the advent of the git caching, using a branch name does not
work anymore. Indeed, we now do a git-fetch, and that does not create
local master branch. So we can't check out master, because it does not
exist locally. And for other branches, as noticed above, the local
branch does not get udpated to the remote one.

Furthermore, the local branches are only created by chance, again as a
side-effect of trying to fetch the "special refs".

So, we can't say that we reliably support the use of a branch name.

Update the manual to state that using a branch does not work. Remove
the 'stable' example, as it looked like the name of a stable branch;
instead, replace it with a version string that ressemble a tag.

Fix the layout of the manual by making the version examples an actual
bulleted list.

Note: the above is only entirely true for git. For Mercurial, CVS and
subversion, the status may be mixed, but nonetheless, using branches is
still a bad idea, if at least because it is not reproducible, and
because Buildroot does not even follow the branch. So, we do not
differentiate between the various SCMs, and just flatly state that using
a branch name is not supported.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Peter Korsgaard <peter@korsgaard.com>

---
Changes v1 -> v2:
  - enhance commit log with additional details  (Ricardo)
  - blurb about
  - typoes in commit log  (Ricardo)
---
 docs/manual/adding-packages-generic.txt | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

Comments

Thomas De Schampheleire May 11, 2018, 7:38 p.m. UTC | #1
2018-05-11 17:50 GMT+02:00 Yann E. MORIN <yann.morin.1998@free.fr>:
> For various reasons, we've always suggested users to avoid using a
> branch as version string for their packages, because it does not work
> as a they would expect:
>
>   - it is not reproducible, because the branch may change between two
>     builds that are done at different times;
>
>   - it does not even follow the branch, as Buildroot anyway generates
>     a local tarball, which it will reuse on subsequent builds.
>     Furthermore, since we fetch and not pull, any existing local branch
>     is not updated.
>
> Yet, until recently, using a branch name would just work (with the
> above limitations): the git tree was cloned, the branch checked out,
> and the tarball created.
>
> But with the advent of the git caching, using a branch name does not
> work anymore. Indeed, we now do a git-fetch, and that does not create
> local master branch. So we can't check out master, because it does not
> exist locally. And for other branches, as noticed above, the local
> branch does not get udpated to the remote one.
>
> Furthermore, the local branches are only created by chance, again as a
> side-effect of trying to fetch the "special refs".
>
> So, we can't say that we reliably support the use of a branch name.
>
> Update the manual to state that using a branch does not work. Remove
> the 'stable' example, as it looked like the name of a stable branch;
> instead, replace it with a version string that ressemble a tag.
>
> Fix the layout of the manual by making the version examples an actual
> bulleted list.
>
> Note: the above is only entirely true for git. For Mercurial, CVS and
> subversion, the status may be mixed, but nonetheless, using branches is
> still a bad idea, if at least because it is not reproducible, and
> because Buildroot does not even follow the branch. So, we do not
> differentiate between the various SCMs, and just flatly state that using
> a branch name is not supported.
>
> Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
> Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
> Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
> Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
> Cc: Peter Korsgaard <peter@korsgaard.com>
>
> ---
> Changes v1 -> v2:
>   - enhance commit log with additional details  (Ricardo)
>   - blurb about
>   - typoes in commit log  (Ricardo)
> ---
>  docs/manual/adding-packages-generic.txt | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/docs/manual/adding-packages-generic.txt b/docs/manual/adding-packages-generic.txt
> index 7e1f246752..8ea8ce6037 100644
> --- a/docs/manual/adding-packages-generic.txt
> +++ b/docs/manual/adding-packages-generic.txt
> @@ -199,12 +199,12 @@ information is (assuming the package name is +libfoo+) :
>  * +LIBFOO_VERSION+, mandatory, must contain the version of the
>    package. Note that if +HOST_LIBFOO_VERSION+ doesn't exist, it is
>    assumed to be the same as +LIBFOO_VERSION+. It can also be a
> -  revision number, branch or tag for packages that are fetched
> -  directly from their revision control system. +
> -  Examples: +
> -    +LIBFOO_VERSION = 0.1.2+ +
> -    +LIBFOO_VERSION = cb9d6aa9429e838f0e54faa3d455bcbab5eef057+ +
> -    +LIBFOO_VERSION = stable+
> +  revision number or a tag for packages that are fetched directly
> +  from their revision control system. Do not use a branch name as

I noticed this is the only place where we say 'revision control
system', other places use 'version control system' which is in line
with the acronym VCS.
I suggest to change this at the same time.

Best regards,
Thomas
Thomas Petazzoni May 11, 2018, 9:05 p.m. UTC | #2
Hello,

On Fri, 11 May 2018 17:50:58 +0200, Yann E. MORIN wrote:
> For various reasons, we've always suggested users to avoid using a
> branch as version string for their packages, because it does not work
> as a they would expect:
> 
>   - it is not reproducible, because the branch may change between two
>     builds that are done at different times;
> 
>   - it does not even follow the branch, as Buildroot anyway generates
>     a local tarball, which it will reuse on subsequent builds.
>     Furthermore, since we fetch and not pull, any existing local branch
>     is not updated.
> 
> Yet, until recently, using a branch name would just work (with the
> above limitations): the git tree was cloned, the branch checked out,
> and the tarball created.
> 
> But with the advent of the git caching, using a branch name does not
> work anymore. Indeed, we now do a git-fetch, and that does not create
> local master branch. So we can't check out master, because it does not
> exist locally. And for other branches, as noticed above, the local
> branch does not get udpated to the remote one.
> 
> Furthermore, the local branches are only created by chance, again as a
> side-effect of trying to fetch the "special refs".
> 
> So, we can't say that we reliably support the use of a branch name.
> 
> Update the manual to state that using a branch does not work. Remove
> the 'stable' example, as it looked like the name of a stable branch;
> instead, replace it with a version string that ressemble a tag.
> 
> Fix the layout of the manual by making the version examples an actual
> bulleted list.
> 
> Note: the above is only entirely true for git. For Mercurial, CVS and
> subversion, the status may be mixed, but nonetheless, using branches is
> still a bad idea, if at least because it is not reproducible, and
> because Buildroot does not even follow the branch. So, we do not
> differentiate between the various SCMs, and just flatly state that using
> a branch name is not supported.
> 
> Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
> Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
> Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
> Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
> Cc: Peter Korsgaard <peter@korsgaard.com>
> 
> ---
> Changes v1 -> v2:
>   - enhance commit log with additional details  (Ricardo)
>   - blurb about
>   - typoes in commit log  (Ricardo)
> ---
>  docs/manual/adding-packages-generic.txt | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)

Applied to master after using "version control system", as suggested by
Thomas DS. Thanks!

Thomas
Peter Korsgaard May 28, 2018, 2:22 p.m. UTC | #3
>>>>> "Yann" == Yann E MORIN <yann.morin.1998@free.fr> writes:

 > For various reasons, we've always suggested users to avoid using a
 > branch as version string for their packages, because it does not work
 > as a they would expect:

 >   - it is not reproducible, because the branch may change between two
 >     builds that are done at different times;

 >   - it does not even follow the branch, as Buildroot anyway generates
 >     a local tarball, which it will reuse on subsequent builds.
 >     Furthermore, since we fetch and not pull, any existing local branch
 >     is not updated.

 > Yet, until recently, using a branch name would just work (with the
 > above limitations): the git tree was cloned, the branch checked out,
 > and the tarball created.

 > But with the advent of the git caching, using a branch name does not
 > work anymore. Indeed, we now do a git-fetch, and that does not create
 > local master branch. So we can't check out master, because it does not
 > exist locally. And for other branches, as noticed above, the local
 > branch does not get udpated to the remote one.

 > Furthermore, the local branches are only created by chance, again as a
 > side-effect of trying to fetch the "special refs".

 > So, we can't say that we reliably support the use of a branch name.

 > Update the manual to state that using a branch does not work. Remove
 > the 'stable' example, as it looked like the name of a stable branch;
 > instead, replace it with a version string that ressemble a tag.

 > Fix the layout of the manual by making the version examples an actual
 > bulleted list.

 > Note: the above is only entirely true for git. For Mercurial, CVS and
 > subversion, the status may be mixed, but nonetheless, using branches is
 > still a bad idea, if at least because it is not reproducible, and
 > because Buildroot does not even follow the branch. So, we do not
 > differentiate between the various SCMs, and just flatly state that using
 > a branch name is not supported.

 > Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
 > Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
 > Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
 > Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
 > Cc: Peter Korsgaard <peter@korsgaard.com>

 > ---
 > Changes v1 -> v2:
 >   - enhance commit log with additional details  (Ricardo)
 >   - blurb about
 >   - typoes in commit log  (Ricardo)

Committed to 2018.02.x, thanks. While we do not have the reworked
download support (and hence the issue that triggered this), we should
still not recommend people to use branch names as versions.
diff mbox series

Patch

diff --git a/docs/manual/adding-packages-generic.txt b/docs/manual/adding-packages-generic.txt
index 7e1f246752..8ea8ce6037 100644
--- a/docs/manual/adding-packages-generic.txt
+++ b/docs/manual/adding-packages-generic.txt
@@ -199,12 +199,12 @@  information is (assuming the package name is +libfoo+) :
 * +LIBFOO_VERSION+, mandatory, must contain the version of the
   package. Note that if +HOST_LIBFOO_VERSION+ doesn't exist, it is
   assumed to be the same as +LIBFOO_VERSION+. It can also be a
-  revision number, branch or tag for packages that are fetched
-  directly from their revision control system. +
-  Examples: +
-    +LIBFOO_VERSION = 0.1.2+ +
-    +LIBFOO_VERSION = cb9d6aa9429e838f0e54faa3d455bcbab5eef057+ +
-    +LIBFOO_VERSION = stable+
+  revision number or a tag for packages that are fetched directly
+  from their revision control system. Do not use a branch name as
+  version; it does not work. Examples:
+  ** a version for a release tarball: +LIBFOO_VERSION = 0.1.2+
+  ** a sha1 for a git tree: +LIBFOO_VERSION = cb9d6aa9429e838f0e54faa3d455bcbab5eef057+
+  ** a tag for a git tree +LIBFOO_VERSION = v0.1.2+
 
 * +LIBFOO_SOURCE+ may contain the name of the tarball of the package,
   which Buildroot will use to download the tarball from