[PATCH 0/5] Reducing system size: locales and initrd

Hello Guix,

I noticed¹ that ‘bare-bones.tmpl’ now leads to instances around

1.5 GiB, which I think is totally unreasonable, to say the least.

This patch series does the easy changes to get it to ~1.1 GiB,

which is still unreasonable, but at least should allow our installation

tests to run without having to increase the disk size again.

Thoughts?

The next but more difficult steps will be stripping Guix and Linux:

Ludo’.

¹ https://ci.guix.gnu.org/build/3708442/details

Ludovic Courtès (5):

system: Remove ‘glibc-2.33’ from ‘%default-locale-libcs’.

system: ‘operating-system-locale-definitions’ includes the OS’ locale.

locale: Shrink ‘%default-locale-definitions’ from 34 to 10 locales.

services: build-vm: Provide only one locale.

linux-initrd: Further strip the static Guile.

doc/guix.texi | 5 +--

gnu/packages/make-bootstrap.scm | 37 ++++++++++++++++++---

gnu/services/virtualization.scm | 9 +++++

gnu/system.scm | 28 ++++++++--------

gnu/system/linux-initrd.scm | 4 +--

gnu/system/locale.scm | 58 +++++++++------------------------

6 files changed, 77 insertions(+), 64 deletions(-)

base-commit: cd45294d576975a3bff2f755764a3f46f09ea6f9

--

2.41.0

The transition from glibc 2.33 to 2.35 was done in

c919bfefd98bf2e29549539b4e28e6dc2a8a6f32 (one year ago), so we can

assume that the backward-compatible locales are no longer needed by

default.

This removes 92 MiB from the system closure.

Change-Id: I85948bbe6b2d424f9f158eeafdb5543688c66c6b

---

gnu/system/locale.scm | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)

thunked. Default to (operating-system-default-locale-definitions os).

(operating-system-default-locale-definitions): New procedure.

(operating-system-locale-directory): Remove logic to add the OS

‘locale’, instead assuming (operating-system-locale-directory os) has

everything.

Change-Id: I8798fa0de7421e0bfc07870cc3cae6b7f99fa47d

---

doc/guix.texi | 3 ++-

gnu/system.scm | 28 +++++++++++++++-------------

2 files changed, 17 insertions(+), 14 deletions(-)

This reduces the default set of locales from 92 MiB to 28 MiB.

locales.

Change-Id: I3c092604301d69db591957bcfd62a062c3ac5ab0

---

gnu/system/locale.scm | 56 ++++++++++++-------------------------------

1 file changed, 15 insertions(+), 41 deletions(-)

‘guile-static-initrd’ weighs in at 46 MiB, compared to 54 MiB for

‘guile-static-stripped’ (15% reduction).

‘directories-to-remove’ parameter and honor it.

(%guile-static-initrd): New variable.

‘%guile-static-initrd’.

Change-Id: I2baf06fed7a3698433e7c83b1d7726054a8c746e

---

doc/guix.texi | 2 +-

gnu/packages/make-bootstrap.scm | 37 ++++++++++++++++++++++++++++-----

gnu/system/linux-initrd.scm | 4 ++--

3 files changed, 35 insertions(+), 8 deletions(-)

[locale, locale-definitions]: New fields.

Change-Id: Ieb24b3a0c210291d8c04412e4c263b5e377b5704

---

gnu/services/virtualization.scm | 9 +++++++++

1 file changed, 9 insertions(+)

Hello Ludo. Even though with my base commit I get 70MB more than you do

(from 1460.5MB down to 1296.1MB), your patches still bring good

improvement.

So (operating-system-locale-definitions os) can now be changed to

something other than operating-system-locale-definitions that does not

include locale. Perhaps this is useful when one does not want utf8 but

EUC-JP like in the “info "(guix)Locales"”? Okay.

I agree with shrinking default locales, but do not understand what

benefit it brings that U.N. languages are always included. Perhaps it

would be more useful to include en_US only? Locales like fr_CA and

ar_JO do not even have separate teams on translationproject.org. There

are reasons for developers to include interesting locales for testing

like tr_TR or az_AZ, where capitalized i is ? and small I is ?, breaking

all kinds of programs, but few people would know.

Regards,

Florian

Hello Ludo, with a clearer mind now:

Ludovic Courtès <ludo@gnu.org> writes:

This is a regression. I have in my /run/current-system/configuration.scm

(locale "de_DE.utf8")

(locale-definitions

(cons* (locale-definition

(name "ar_SY.utf8") (source "ar_SY"))

(locale-definition

(name "bn_BD.utf8") (source "bn_BD"))

(locale-definition

(name "hu_HU.utf8") (source "hu_HU"))

(locale-definition

(name "ka_GE.utf8") (source "ka_GE"))

(locale-definition

(name "lt_LT.utf8") (source "lt_LT"))

(locale-definition

(name "oc_FR.utf8") (source "oc_FR"))

(locale-definition

(name "pt_BR.utf8") (source "pt_BR"))

(locale-definition

(name "si_LK.utf8") (source "si_LK"))

(locale-definition

(name "sk_SK.utf8") (source "sk_SK"))

(locale-definition

(name "sr_RS.utf8") (source "sr_RS"))

%default-locale-definitions))

but de_DE.utf8 is not available and

“export LC_ALL=de_DE.utf8” gives an error.

scheme@(guix-user)> ,use (gnu system locale)

scheme@(guix-user)> %default-locale-definitions

$1 = (#<<locale-definition> name: "ar_DZ.utf8" source: "ar_DZ" charset: "UTF-8"> #<<locale-definition> name: "ar_JO.utf8" source: "ar_JO" charset: "UTF-8"> #<<locale-definition> name: "en_GB.utf8" source: "en_GB" charset: "UTF-8"> #<<locale-definition> name: "en_US.utf8" source: "en_US" charset: "UTF-8"> #<<locale-definition> name: "es_AR.utf8" source: "es_AR" charset: "UTF-8"> #<<locale-definition> name: "es_ES.utf8" source: "es_ES" charset: "UTF-8"> #<<locale-definition> name: "fr_CA.utf8" source: "fr_CA" charset: "UTF-8"> #<<locale-definition> name: "fr_FR.utf8" source: "fr_FR" charset: "UTF-8"> #<<locale-definition> name: "ru_RU.utf8" source: "ru_RU" charset: "UTF-8"> #<<locale-definition> name: "zh_CN.utf8" source: "zh_CN" charset: "UTF-8">)

Regards,

Florian

"pelzflorian (Florian Pelz)" <pelzflorian@pelzflorian.de> writes:

I meant to also write:

scheme@(guix-user)> ,use (gnu system)

scheme@(guix-user)> (operating-system-locale-definitions (load "/run/current-system/configuration.scm"))

$2 = (#<<locale-definition> name: "ar_SY.utf8" source: "ar_SY" charset: "UTF-8"> #<<locale-definition> name: "bn_BD.utf8" source: "bn_BD" charset: "UTF-8"> #<<locale-definition> name: "hu_HU.utf8" source: "hu_HU" charset: "UTF-8"> #<<locale-definition> name: "ka_GE.utf8" source: "ka_GE" charset: "UTF-8"> #<<locale-definition> name: "lt_LT.utf8" source: "lt_LT" charset: "UTF-8"> #<<locale-definition> name: "oc_FR.utf8" source: "oc_FR" charset: "UTF-8"> #<<locale-definition> name: "pt_BR.utf8" source: "pt_BR" charset: "UTF-8"> #<<locale-definition> name: "si_LK.utf8" source: "si_LK" charset: "UTF-8"> #<<locale-definition> name: "sk_SK.utf8" source: "sk_SK" charset: "UTF-8"> #<<locale-definition> name: "sr_RS.utf8" source: "sr_RS" charset: "UTF-8"> #<<locale-definition> name: "ar_DZ.utf8" source: "ar_DZ" charset: "UTF-8"> #<<locale-definition> name: "ar_JO.utf8" source: "ar_JO" charset: "UTF-8"> #<<locale-definition> name: "en_GB.utf8" source: "en_GB" charset: "UTF-8"> #<<locale-definition> name: "en_US.utf8" source: "en_US" charset: "UTF-8"> #<<locale-definition> name: "es_AR.utf8" source: "es_AR" charset: "UTF-8"> #<<locale-definition> name: "es_ES.utf8" source: "es_ES" charset: "UTF-8"> #<<locale-definition> name: "fr_CA.utf8" source: "fr_CA" charset: "UTF-8"> #<<locale-definition> name: "fr_FR.utf8" source: "fr_FR" charset: "UTF-8"> #<<locale-definition> name: "ru_RU.utf8" source: "ru_RU" charset: "UTF-8"> #<<locale-definition> name: "zh_CN.utf8" source: "zh_CN" charset: "UTF-8">)

No de_DE in there, but it should be.

Regards,

Florian

Guten Tag Florian,

"pelzflorian (Florian Pelz)" <pelzflorian@pelzflorian.de> skribis:

My idea was to include by default a few locales that people may

“typically” find useful, in addition to their chosen one.

Of course, it’s an arbitrary choice (choosing the six UN languages makes

it slightly less arbitrary).

My goal is *not* to provide a default set useful for developers

(otherwise German, Greek, Turkish, and non-UTF-8 locales would be

welcome).

Now, should we remove fr_CA and ar_JO? Maybe! I’m open to really any

changes in this list; I just think it should go beyond en_US.

Thoughts?

Ludo’.

"pelzflorian (Florian Pelz)" <pelzflorian@pelzflorian.de> skribis:

Right.

I would call it a semantic change rather than a regression (a feature

rather than a bug :-)).

It’s more in line with what we’re doing elsewhere: the default is to

include ‘locale’ in the locale definitions, but we let the user override

that if they wish.

Perhaps we should at least clarify the manual?

Ludo’.

Bonjour Ludo. :)

Ludovic Courtès <ludo@gnu.org> writes:

Hmm but how would they use it? Ohh I just noticed that these languages

configured here will be the options in the GNOME Settings. I do agree

now that the six languages are a standard set, although Russian has less

speakers than Portuguese [1], which gets discussed [2] and I would tend

to also include Portuguese then, because it has translations unlike

Bengali or Hindi.

I would remove ar_JO. Probably also remove fr_CA, because someone from

Canada will configure fr_CA during Guix installation.

Regards,

Florian

[1]

[2]

Ludovic Courtès <ludo@gnu.org> writes:

It is a regression when with the same OS config.scm, my GNOME is no

longer in German. Perhaps I’m missing something, but the only gain from

making it configurable to not include the configured default locale in

locale-definitions would be that it becomes possible to not include the

default locale language’s utf8 encoding, but utf8 is good for everyone.

I would have expected the prior behavior, i.e. the setting in locale to

be available regardless of locale-definitions. To me it does not seem

necessary to be documented explicitly, but maybe better be explicit.

Regards,

Florian

"pelzflorian (Florian Pelz)" <pelzflorian@pelzflorian.de> skribis:

[...]

OK, let’s drop this patch then.

I was focusing on interface consistency but in this particular case what

you describe suggests interface consistency is a secondary concern.

Thanks,

Ludo’.

"pelzflorian (Florian Pelz)" <pelzflorian@pelzflorian.de> skribis:

[...]

I don’t use a desktop environment; my use case would be:

export LC_ALL=…

or, say:

LANGUAGE=fr:es:en

Yeah, the UN language choice is base on imperialism more than

popularity.

OK, let’s drop these two and add pt_BR and pt_PT then.

Thanks!

Ludo’.