| Message ID | 20200614093330.17516-1-mail@aparcar.org |
|---|---|
| State | Not Applicable |
| Headers | show |
| Series | [OpenWrt-Devel,RFC] build: disable target name in image filename | expand |
On Sat, Jun 13, 2020 at 11:33:31PM -1000, Paul Spooren wrote: > The target/subtarget information in image filenames is barely of any use > for developers or end users. > > A developer reads the profile name and the target is either obvious due > to previous work or `cd targets/ && grep -r <profile>` tells the target > within 3ms. If no buildroot is available `cat <image> | tail -c 200` > allows a look at the attached metadata which includes the > target/subtarget. > > For users the information is entirely useless and maybe even harmful. > Target names like `cortexa9` could easily be mistaken as an actual > device name while the only relevant information would be > `linksys_wrt3200acm`. Images are more realistically downloaded via a > Wiki entry or a firmware wizard. > > This commit therefore adds the new image option called > CONFIG_TARGET_FILENAMES to make the target/subtarget filename part > optional. It is disabled by default. > > As the profile name `generic` appears multiple times in the x86 target > as well as in oceton and ath25, the proposed patch on GitHub should be > merged first: > * treewide: use unique profile names #3082 > https://github.com/openwrt/openwrt/pull/3082 > > Newly produced files would look like the following: > * openwrt-linksys_wrt3200acm-initramfs-kernel.bin > * openwrt-linksys_wrt3200acm.manifest > * openwrt-linksys_wrt3200acm-squashfs-factory.img > * openwrt-linksys_wrt3200acm-squashfs-sysupgrade.bin > > Signed-off-by: Paul Spooren <mail@aparcar.org> Once all the requirements (unique and informative image names) are in place to make this work: Acked-by: Daniel Golle <daniel@makrotopia.org> > --- > It's been a while since I made a controversial patch[0] so it feels > about time. > > [0]: https://github.com/openwrt/openwrt/pull/2107 > > include/image.mk | 9 +++++---- > package/base-files/image-config.in | 9 +++++++++ > 2 files changed, 14 insertions(+), 4 deletions(-) > > diff --git a/include/image.mk b/include/image.mk > index 984b64fb9c..c6fc467c9e 100644 > --- a/include/image.mk > +++ b/include/image.mk > @@ -37,11 +37,12 @@ KDIR=$(KERNEL_BUILD_DIR) > KDIR_TMP=$(KDIR)/tmp > DTS_DIR:=$(LINUX_DIR)/arch/$(LINUX_KARCH)/boot/dts > > +IMG_PREFIX_TARGET:=$(if $(CONFIG_TARGET_FILENAMES),$(BOARD)$(if $(SUBTARGET),-$(SUBTARGET))-) > IMG_PREFIX_EXTRA:=$(if $(EXTRA_IMAGE_NAME),$(call sanitize,$(EXTRA_IMAGE_NAME))-) > IMG_PREFIX_VERNUM:=$(if $(CONFIG_VERSION_FILENAMES),$(call sanitize,$(VERSION_NUMBER))-) > IMG_PREFIX_VERCODE:=$(if $(CONFIG_VERSION_CODE_FILENAMES),$(call sanitize,$(VERSION_CODE))-) > > -IMG_PREFIX:=$(VERSION_DIST_SANITIZED)-$(IMG_PREFIX_VERNUM)$(IMG_PREFIX_VERCODE)$(IMG_PREFIX_EXTRA)$(BOARD)$(if $(SUBTARGET),-$(SUBTARGET)) > +IMG_PREFIX:=$(VERSION_DIST_SANITIZED)-$(IMG_PREFIX_VERNUM)$(IMG_PREFIX_VERCODE)$(IMG_PREFIX_EXTRA)$(IMG_PREFIX_TARGET) > IMG_ROOTFS:=$(IMG_PREFIX)-rootfs > IMG_COMBINED:=$(IMG_PREFIX)-combined > IMG_PART_SIGNATURE:=$(shell echo $(SOURCE_DATE_EPOCH)$(LINUX_VERMAGIC) | mkhash md5 | cut -b1-8) > @@ -293,7 +294,7 @@ endef > > define Image/Manifest > $(call opkg,$(TARGET_DIR_ORIG)) list-installed > \ > - $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),-$(PROFILE_SANITIZED)).manifest > + $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),$(PROFILE_SANITIZED)).manifest > endef > > define Image/gzip-ext4-padded-squashfs > @@ -317,7 +318,7 @@ ifdef CONFIG_TARGET_ROOTFS_TARGZ > define Image/Build/targz > $(TAR) -cp --numeric-owner --owner=0 --group=0 --mode=a-s --sort=name \ > $(if $(SOURCE_DATE_EPOCH),--mtime="@$(SOURCE_DATE_EPOCH)") \ > - -C $(TARGET_DIR)/ . | gzip -9n > $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),-$(PROFILE_SANITIZED))-rootfs.tar.gz > + -C $(TARGET_DIR)/ . | gzip -9n > $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),$(PROFILE_SANITIZED))-rootfs.tar.gz > endef > endif > > @@ -385,7 +386,7 @@ define Device/Init > > IMAGES := > ARTIFACTS := > - IMAGE_PREFIX := $(IMG_PREFIX)-$(1) > + IMAGE_PREFIX := $(IMG_PREFIX)$(1) > IMAGE_NAME = $$(IMAGE_PREFIX)-$$(1)-$$(2) > IMAGE_SIZE := > KERNEL_PREFIX = $$(IMAGE_PREFIX) > diff --git a/package/base-files/image-config.in b/package/base-files/image-config.in > index 3432db525a..5a70d51a7a 100644 > --- a/package/base-files/image-config.in > +++ b/package/base-files/image-config.in > @@ -264,6 +264,15 @@ if VERSIONOPT > Enable this to include the revision identifier or the configured > version code into the firmware image, SDK- and Image Builder archive > file names > + > + config TARGET_FILENAMES > + bool > + prompt "Target and subtarget in filenames" > + default n > + help > + Enable this to include the target (and subtarget) in > + firmware image, SDK- and Image Builder archive file names > + > endif > > > -- > 2.20.1 > > > _______________________________________________ > openwrt-devel mailing list > openwrt-devel@lists.openwrt.org > https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Hi, > -----Original Message----- > From: openwrt-devel [mailto:openwrt-devel-bounces@lists.openwrt.org] > On Behalf Of Paul Spooren > Sent: Sonntag, 14. Juni 2020 11:34 > To: openwrt-devel@lists.openwrt.org > Subject: [OpenWrt-Devel] [PATCH][RFC] build: disable target name in image > filename > > The target/subtarget information in image filenames is barely of any use for > developers or end users. > > A developer reads the profile name and the target is either obvious due to > previous work or `cd targets/ && grep -r <profile>` tells the target within > 3ms. If no buildroot is available `cat <image> | tail -c 200` allows a look at the > attached metadata which includes the target/subtarget. > > For users the information is entirely useless and maybe even harmful. > Target names like `cortexa9` could easily be mistaken as an actual device > name while the only relevant information would be `linksys_wrt3200acm`. > Images are more realistically downloaded via a Wiki entry or a firmware > wizard. > > This commit therefore adds the new image option called > CONFIG_TARGET_FILENAMES to make the target/subtarget filename part > optional. It is disabled by default. > > As the profile name `generic` appears multiple times in the x86 target as well > as in oceton and ath25, the proposed patch on GitHub should be merged > first: > * treewide: use unique profile names #3082 > https://github.com/openwrt/openwrt/pull/3082 > > Newly produced files would look like the following: > * openwrt-linksys_wrt3200acm-initramfs-kernel.bin > * openwrt-linksys_wrt3200acm.manifest > * openwrt-linksys_wrt3200acm-squashfs-factory.img > * openwrt-linksys_wrt3200acm-squashfs-sysupgrade.bin I just think of ar71xx and ath79, where we have the same device but different targets. Of course, the name won't be exactly equal, as ath79 will have e.g. tplink_ prefix and ar71xx won't. For bcm63xx, we have two subtargets that build the same devices. If we look at PR#2957, we might have a now bmips target at some point that features the same devices as bcm63xx. This won't necessarily break anything, as images will still be in different folders (at least in /bin). However, we couldn't tell the difference between ar71xx/ath79 or similar from the image name (easily) after this change, or whether it's generic or smp for bcm63xx. For my personal taste, this drawback is bigger that the gain we will get from removing the target/subtarget part. So, unless there is overwhelming support, I tend to NAK this. Best Adrian > > Signed-off-by: Paul Spooren <mail@aparcar.org> > --- > It's been a while since I made a controversial patch[0] so it feels about time. > > [0]: https://github.com/openwrt/openwrt/pull/2107 > > include/image.mk | 9 +++++---- > package/base-files/image-config.in | 9 +++++++++ > 2 files changed, 14 insertions(+), 4 deletions(-) > > diff --git a/include/image.mk b/include/image.mk index > 984b64fb9c..c6fc467c9e 100644 > --- a/include/image.mk > +++ b/include/image.mk > @@ -37,11 +37,12 @@ KDIR=$(KERNEL_BUILD_DIR) > KDIR_TMP=$(KDIR)/tmp > DTS_DIR:=$(LINUX_DIR)/arch/$(LINUX_KARCH)/boot/dts > > +IMG_PREFIX_TARGET:=$(if $(CONFIG_TARGET_FILENAMES),$(BOARD)$(if > +$(SUBTARGET),-$(SUBTARGET))-) > IMG_PREFIX_EXTRA:=$(if $(EXTRA_IMAGE_NAME),$(call > sanitize,$(EXTRA_IMAGE_NAME))-) IMG_PREFIX_VERNUM:=$(if > $(CONFIG_VERSION_FILENAMES),$(call sanitize,$(VERSION_NUMBER))-) > IMG_PREFIX_VERCODE:=$(if $(CONFIG_VERSION_CODE_FILENAMES),$(call > sanitize,$(VERSION_CODE))-) > > -IMG_PREFIX:=$(VERSION_DIST_SANITIZED)- > $(IMG_PREFIX_VERNUM)$(IMG_PREFIX_VERCODE)$(IMG_PREFIX_EXTRA)$ > (BOARD)$(if $(SUBTARGET),-$(SUBTARGET)) > +IMG_PREFIX:=$(VERSION_DIST_SANITIZED)- > $(IMG_PREFIX_VERNUM)$(IMG_PREFIX_ > +VERCODE)$(IMG_PREFIX_EXTRA)$(IMG_PREFIX_TARGET) > IMG_ROOTFS:=$(IMG_PREFIX)-rootfs > IMG_COMBINED:=$(IMG_PREFIX)-combined > IMG_PART_SIGNATURE:=$(shell echo > $(SOURCE_DATE_EPOCH)$(LINUX_VERMAGIC) | mkhash md5 | cut -b1-8) > @@ -293,7 +294,7 @@ endef > > define Image/Manifest > $(call opkg,$(TARGET_DIR_ORIG)) list-installed > \ > - $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),- > $(PROFILE_SANITIZED)).manifest > + $(BIN_DIR)/$(IMG_PREFIX)$(if > +$(PROFILE_SANITIZED),$(PROFILE_SANITIZED)).manifest > endef > > define Image/gzip-ext4-padded-squashfs > @@ -317,7 +318,7 @@ ifdef CONFIG_TARGET_ROOTFS_TARGZ > define Image/Build/targz > $(TAR) -cp --numeric-owner --owner=0 --group=0 --mode=a-s -- > sort=name \ > $(if $(SOURCE_DATE_EPOCH),-- > mtime="@$(SOURCE_DATE_EPOCH)") \ > - -C $(TARGET_DIR)/ . | gzip -9n > > $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),- > $(PROFILE_SANITIZED))-rootfs.tar.gz > + -C $(TARGET_DIR)/ . | gzip -9n > > $(BIN_DIR)/$(IMG_PREFIX)$(if > +$(PROFILE_SANITIZED),$(PROFILE_SANITIZED))-rootfs.tar.gz > endef > endif > > @@ -385,7 +386,7 @@ define Device/Init > > IMAGES := > ARTIFACTS := > - IMAGE_PREFIX := $(IMG_PREFIX)-$(1) > + IMAGE_PREFIX := $(IMG_PREFIX)$(1) > IMAGE_NAME = $$(IMAGE_PREFIX)-$$(1)-$$(2) > IMAGE_SIZE := > KERNEL_PREFIX = $$(IMAGE_PREFIX) > diff --git a/package/base-files/image-config.in b/package/base-files/image- > config.in > index 3432db525a..5a70d51a7a 100644 > --- a/package/base-files/image-config.in > +++ b/package/base-files/image-config.in > @@ -264,6 +264,15 @@ if VERSIONOPT > Enable this to include the revision identifier or the > configured > version code into the firmware image, SDK- and > Image Builder archive > file names > + > + config TARGET_FILENAMES > + bool > + prompt "Target and subtarget in filenames" > + default n > + help > + Enable this to include the target (and subtarget) in > + firmware image, SDK- and Image Builder archive file > names > + > endif > > > -- > 2.20.1 > > > _______________________________________________ > openwrt-devel mailing list > openwrt-devel@lists.openwrt.org > https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Hi, On 6/14/20 10:00 AM, mail@adrianschmutzler.de wrote: >> -----Original Message----- >> From: openwrt-devel [mailto:openwrt-devel-bounces@lists.openwrt.org] >> On Behalf Of Paul Spooren >> Sent: Sonntag, 14. Juni 2020 11:34 >> To: openwrt-devel@lists.openwrt.org >> Subject: [OpenWrt-Devel] [PATCH][RFC] build: disable target name in image >> filename >> >> The target/subtarget information in image filenames is barely of any use for >> developers or end users. >> >> A developer reads the profile name and the target is either obvious due to >> previous work or `cd targets/ && grep -r <profile>` tells the target within >> 3ms. If no buildroot is available `cat <image> | tail -c 200` allows a look at the >> attached metadata which includes the target/subtarget. >> >> For users the information is entirely useless and maybe even harmful. >> Target names like `cortexa9` could easily be mistaken as an actual device >> name while the only relevant information would be `linksys_wrt3200acm`. >> Images are more realistically downloaded via a Wiki entry or a firmware >> wizard. >> >> This commit therefore adds the new image option called >> CONFIG_TARGET_FILENAMES to make the target/subtarget filename part >> optional. It is disabled by default. >> >> As the profile name `generic` appears multiple times in the x86 target as well >> as in oceton and ath25, the proposed patch on GitHub should be merged >> first: >> * treewide: use unique profile names #3082 >> https://github.com/openwrt/openwrt/pull/3082 >> >> Newly produced files would look like the following: >> * openwrt-linksys_wrt3200acm-initramfs-kernel.bin >> * openwrt-linksys_wrt3200acm.manifest >> * openwrt-linksys_wrt3200acm-squashfs-factory.img >> * openwrt-linksys_wrt3200acm-squashfs-sysupgrade.bin > I just think of ar71xx and ath79, where we have the same device but different targets. Of course, the name won't be exactly equal, as ath79 will have e.g. tplink_ prefix and ar71xx won't. Isn't ar71xx removed from master builds? It's neither at snapshot https://downloads.openwrt.org/snapshots/targets/ nor planed to be re-added to any upcoming release, is it? > For bcm63xx, we have two subtargets that build the same devices. > If we look at PR#2957, we might have a now bmips target at some point that features the same devices as bcm63xx. Can you please explain why that's the case? Why do we offer different images for the same device? I understand that for ar71xx -> ath79 within a transfer period but it's never the scope to offer different "flavors" long term, is it? > This won't necessarily break anything, as images will still be in different folders (at least in /bin). I would be at least confusing and reverts the "unique profile name" idea. > However, we couldn't tell the difference between ar71xx/ath79 or similar from the image name (easily) after this change, or whether it's generic or smp for bcm63xx. For my personal taste, this drawback is bigger that the gain we will get from removing the target/subtarget part. Again, this sounds like a undesirable state where we not only build but also maintain multiple images for the save device. Wouldn't it be possible to add the target to all those "legacy images", however remove it wherever a target uses device tree and images.mk aka has long term support? > So, unless there is overwhelming support, I tend to NAK this. :( > Best Paul > >> Signed-off-by: Paul Spooren <mail@aparcar.org> >> --- >> It's been a while since I made a controversial patch[0] so it feels about time. >> >> [0]: https://github.com/openwrt/openwrt/pull/2107 >> >> include/image.mk | 9 +++++---- >> package/base-files/image-config.in | 9 +++++++++ >> 2 files changed, 14 insertions(+), 4 deletions(-) >> >> diff --git a/include/image.mk b/include/image.mk index >> 984b64fb9c..c6fc467c9e 100644 >> --- a/include/image.mk >> +++ b/include/image.mk >> @@ -37,11 +37,12 @@ KDIR=$(KERNEL_BUILD_DIR) >> KDIR_TMP=$(KDIR)/tmp >> DTS_DIR:=$(LINUX_DIR)/arch/$(LINUX_KARCH)/boot/dts >> >> +IMG_PREFIX_TARGET:=$(if $(CONFIG_TARGET_FILENAMES),$(BOARD)$(if >> +$(SUBTARGET),-$(SUBTARGET))-) >> IMG_PREFIX_EXTRA:=$(if $(EXTRA_IMAGE_NAME),$(call >> sanitize,$(EXTRA_IMAGE_NAME))-) IMG_PREFIX_VERNUM:=$(if >> $(CONFIG_VERSION_FILENAMES),$(call sanitize,$(VERSION_NUMBER))-) >> IMG_PREFIX_VERCODE:=$(if $(CONFIG_VERSION_CODE_FILENAMES),$(call >> sanitize,$(VERSION_CODE))-) >> >> -IMG_PREFIX:=$(VERSION_DIST_SANITIZED)- >> $(IMG_PREFIX_VERNUM)$(IMG_PREFIX_VERCODE)$(IMG_PREFIX_EXTRA)$ >> (BOARD)$(if $(SUBTARGET),-$(SUBTARGET)) >> +IMG_PREFIX:=$(VERSION_DIST_SANITIZED)- >> $(IMG_PREFIX_VERNUM)$(IMG_PREFIX_ >> +VERCODE)$(IMG_PREFIX_EXTRA)$(IMG_PREFIX_TARGET) >> IMG_ROOTFS:=$(IMG_PREFIX)-rootfs >> IMG_COMBINED:=$(IMG_PREFIX)-combined >> IMG_PART_SIGNATURE:=$(shell echo >> $(SOURCE_DATE_EPOCH)$(LINUX_VERMAGIC) | mkhash md5 | cut -b1-8) >> @@ -293,7 +294,7 @@ endef >> >> define Image/Manifest >> $(call opkg,$(TARGET_DIR_ORIG)) list-installed > \ >> - $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),- >> $(PROFILE_SANITIZED)).manifest >> + $(BIN_DIR)/$(IMG_PREFIX)$(if >> +$(PROFILE_SANITIZED),$(PROFILE_SANITIZED)).manifest >> endef >> >> define Image/gzip-ext4-padded-squashfs >> @@ -317,7 +318,7 @@ ifdef CONFIG_TARGET_ROOTFS_TARGZ >> define Image/Build/targz >> $(TAR) -cp --numeric-owner --owner=0 --group=0 --mode=a-s -- >> sort=name \ >> $(if $(SOURCE_DATE_EPOCH),-- >> mtime="@$(SOURCE_DATE_EPOCH)") \ >> - -C $(TARGET_DIR)/ . | gzip -9n > >> $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),- >> $(PROFILE_SANITIZED))-rootfs.tar.gz >> + -C $(TARGET_DIR)/ . | gzip -9n > >> $(BIN_DIR)/$(IMG_PREFIX)$(if >> +$(PROFILE_SANITIZED),$(PROFILE_SANITIZED))-rootfs.tar.gz >> endef >> endif >> >> @@ -385,7 +386,7 @@ define Device/Init >> >> IMAGES := >> ARTIFACTS := >> - IMAGE_PREFIX := $(IMG_PREFIX)-$(1) >> + IMAGE_PREFIX := $(IMG_PREFIX)$(1) >> IMAGE_NAME = $$(IMAGE_PREFIX)-$$(1)-$$(2) >> IMAGE_SIZE := >> KERNEL_PREFIX = $$(IMAGE_PREFIX) >> diff --git a/package/base-files/image-config.in b/package/base-files/image- >> config.in >> index 3432db525a..5a70d51a7a 100644 >> --- a/package/base-files/image-config.in >> +++ b/package/base-files/image-config.in >> @@ -264,6 +264,15 @@ if VERSIONOPT >> Enable this to include the revision identifier or the >> configured >> version code into the firmware image, SDK- and >> Image Builder archive >> file names >> + >> + config TARGET_FILENAMES >> + bool >> + prompt "Target and subtarget in filenames" >> + default n >> + help >> + Enable this to include the target (and subtarget) in >> + firmware image, SDK- and Image Builder archive file >> names >> + >> endif >> >> >> -- >> 2.20.1 >> >> >> _______________________________________________ >> openwrt-devel mailing list >> openwrt-devel@lists.openwrt.org >> https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Hi, > -----Original Message----- > From: openwrt-devel [mailto:openwrt-devel-bounces@lists.openwrt.org] > On Behalf Of Paul Spooren > Sent: Sonntag, 14. Juni 2020 22:22 > To: mail@adrianschmutzler.de; openwrt-devel@lists.openwrt.org > Subject: Re: [OpenWrt-Devel] [PATCH][RFC] build: disable target name in > image filename > > Hi, > > On 6/14/20 10:00 AM, mail@adrianschmutzler.de wrote: > >> -----Original Message----- > >> From: openwrt-devel [mailto:openwrt-devel- > bounces@lists.openwrt.org] > >> On Behalf Of Paul Spooren > >> Sent: Sonntag, 14. Juni 2020 11:34 > >> To: openwrt-devel@lists.openwrt.org > >> Subject: [OpenWrt-Devel] [PATCH][RFC] build: disable target name in > >> image filename > >> > >> The target/subtarget information in image filenames is barely of any > >> use for developers or end users. > >> > >> A developer reads the profile name and the target is either obvious > >> due to previous work or `cd targets/ && grep -r <profile>` tells the > >> target within 3ms. If no buildroot is available `cat <image> | tail > >> -c 200` allows a look at the attached metadata which includes the > target/subtarget. > >> > >> For users the information is entirely useless and maybe even harmful. > >> Target names like `cortexa9` could easily be mistaken as an actual > >> device name while the only relevant information would be > `linksys_wrt3200acm`. > >> Images are more realistically downloaded via a Wiki entry or a > >> firmware wizard. > >> > >> This commit therefore adds the new image option called > >> CONFIG_TARGET_FILENAMES to make the target/subtarget filename part > >> optional. It is disabled by default. > >> > >> As the profile name `generic` appears multiple times in the x86 > >> target as well as in oceton and ath25, the proposed patch on GitHub > >> should be merged > >> first: > >> * treewide: use unique profile names #3082 > >> https://github.com/openwrt/openwrt/pull/3082 > >> > >> Newly produced files would look like the following: > >> * openwrt-linksys_wrt3200acm-initramfs-kernel.bin > >> * openwrt-linksys_wrt3200acm.manifest > >> * openwrt-linksys_wrt3200acm-squashfs-factory.img > >> * openwrt-linksys_wrt3200acm-squashfs-sysupgrade.bin > > I just think of ar71xx and ath79, where we have the same device but > different targets. Of course, the name won't be exactly equal, as ath79 will > have e.g. tplink_ prefix and ar71xx won't. > Isn't ar71xx removed from master builds? It's neither at snapshot > https://downloads.openwrt.org/snapshots/targets/ nor planed to be re- > added to any upcoming release, is it? Yes, but it's just an example for a similar situation which might arise in the future. Then, we even might not have the current situation with the different device names, but may end up with completely identical names except for the target. > > For bcm63xx, we have two subtargets that build the same devices. > > If we look at PR#2957, we might have a now bmips target at some point > that features the same devices as bcm63xx. > Can you please explain why that's the case? Why do we offer different > images for the same device? I understand that for ar71xx -> ath79 within a I don't have any idea why this situation at bcm63xx exists; I just got aware of it at some point. Maybe Noltari or KanjiMonster can help ... > transfer period but it's never the scope to offer different "flavors" > long term, is it? > > This won't necessarily break anything, as images will still be in different > folders (at least in /bin). > I would be at least confusing and reverts the "unique profile name" idea. > > However, we couldn't tell the difference between ar71xx/ath79 or similar > from the image name (easily) after this change, or whether it's generic or > smp for bcm63xx. For my personal taste, this drawback is bigger that the gain > we will get from removing the target/subtarget part. > Again, this sounds like a undesirable state where we not only build but also > maintain multiple images for the save device. Wouldn't it be possible to add > the target to all those "legacy images", however remove it wherever a target > uses device tree and images.mk aka has long term support? Well, just look at the situation in 19.07. There we have both ar71xx and ath79 for the same devices, and even if we wanted, it would actually be quite hard to really filter out the ath79 devices in ar71xx. I really don't think removing the target from image names will pay out in the future. > > So, unless there is overwhelming support, I tend to NAK this. > :( A compromise could be found by just removing the subtarget, but keeping the target in file names. This would mostly solve your problem with the generic names (at least there would be less duplicate info), but there would be significantly less situations where this was an impediment. Normally, no duplicate devices in a target exist, and if they are moved between subtargets, they are actually moved and not copied. The only remaining problem I can think of at the moment would be the bcm63xx situation, and maybe that one can be resolved at low cost. Best Adrian > > Best > Paul > > > >> Signed-off-by: Paul Spooren <mail@aparcar.org> > >> --- > >> It's been a while since I made a controversial patch[0] so it feels about > time. > >> > >> [0]: https://github.com/openwrt/openwrt/pull/2107 > >> > >> include/image.mk | 9 +++++---- > >> package/base-files/image-config.in | 9 +++++++++ > >> 2 files changed, 14 insertions(+), 4 deletions(-) > >> > >> diff --git a/include/image.mk b/include/image.mk index > >> 984b64fb9c..c6fc467c9e 100644 > >> --- a/include/image.mk > >> +++ b/include/image.mk > >> @@ -37,11 +37,12 @@ KDIR=$(KERNEL_BUILD_DIR) > KDIR_TMP=$(KDIR)/tmp > >> DTS_DIR:=$(LINUX_DIR)/arch/$(LINUX_KARCH)/boot/dts > >> > >> +IMG_PREFIX_TARGET:=$(if > $(CONFIG_TARGET_FILENAMES),$(BOARD)$(if > >> +$(SUBTARGET),-$(SUBTARGET))-) > >> IMG_PREFIX_EXTRA:=$(if $(EXTRA_IMAGE_NAME),$(call > >> sanitize,$(EXTRA_IMAGE_NAME))-) IMG_PREFIX_VERNUM:=$(if > >> $(CONFIG_VERSION_FILENAMES),$(call sanitize,$(VERSION_NUMBER))-) > >> IMG_PREFIX_VERCODE:=$(if > $(CONFIG_VERSION_CODE_FILENAMES),$(call > >> sanitize,$(VERSION_CODE))-) > >> > >> -IMG_PREFIX:=$(VERSION_DIST_SANITIZED)- > >> > $(IMG_PREFIX_VERNUM)$(IMG_PREFIX_VERCODE)$(IMG_PREFIX_EXTRA)$ > >> (BOARD)$(if $(SUBTARGET),-$(SUBTARGET)) > >> +IMG_PREFIX:=$(VERSION_DIST_SANITIZED)- > >> $(IMG_PREFIX_VERNUM)$(IMG_PREFIX_ > >> +VERCODE)$(IMG_PREFIX_EXTRA)$(IMG_PREFIX_TARGET) > >> IMG_ROOTFS:=$(IMG_PREFIX)-rootfs > >> IMG_COMBINED:=$(IMG_PREFIX)-combined > >> IMG_PART_SIGNATURE:=$(shell echo > >> $(SOURCE_DATE_EPOCH)$(LINUX_VERMAGIC) | mkhash md5 | cut -b1-8) > @@ > >> -293,7 +294,7 @@ endef > >> > >> define Image/Manifest > >> $(call opkg,$(TARGET_DIR_ORIG)) list-installed > \ > >> - $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),- > >> $(PROFILE_SANITIZED)).manifest > >> + $(BIN_DIR)/$(IMG_PREFIX)$(if > >> +$(PROFILE_SANITIZED),$(PROFILE_SANITIZED)).manifest > >> endef > >> > >> define Image/gzip-ext4-padded-squashfs @@ -317,7 +318,7 @@ ifdef > >> CONFIG_TARGET_ROOTFS_TARGZ > >> define Image/Build/targz > >> $(TAR) -cp --numeric-owner --owner=0 --group=0 --mode=a-s -- > >> sort=name \ > >> $(if $(SOURCE_DATE_EPOCH),-- > >> mtime="@$(SOURCE_DATE_EPOCH)") \ > >> - -C $(TARGET_DIR)/ . | gzip -9n > > >> $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),- > >> $(PROFILE_SANITIZED))-rootfs.tar.gz > >> + -C $(TARGET_DIR)/ . | gzip -9n > > >> $(BIN_DIR)/$(IMG_PREFIX)$(if > >> +$(PROFILE_SANITIZED),$(PROFILE_SANITIZED))-rootfs.tar.gz > >> endef > >> endif > >> > >> @@ -385,7 +386,7 @@ define Device/Init > >> > >> IMAGES := > >> ARTIFACTS := > >> - IMAGE_PREFIX := $(IMG_PREFIX)-$(1) > >> + IMAGE_PREFIX := $(IMG_PREFIX)$(1) > >> IMAGE_NAME = $$(IMAGE_PREFIX)-$$(1)-$$(2) > >> IMAGE_SIZE := > >> KERNEL_PREFIX = $$(IMAGE_PREFIX) > >> diff --git a/package/base-files/image-config.in > >> b/package/base-files/image- config.in index 3432db525a..5a70d51a7a > >> 100644 > >> --- a/package/base-files/image-config.in > >> +++ b/package/base-files/image-config.in > >> @@ -264,6 +264,15 @@ if VERSIONOPT > >> Enable this to include the revision identifier or the > configured > >> version code into the firmware image, SDK- and > Image Builder > >> archive > >> file names > >> + > >> + config TARGET_FILENAMES > >> + bool > >> + prompt "Target and subtarget in filenames" > >> + default n > >> + help > >> + Enable this to include the target (and subtarget) in > >> + firmware image, SDK- and Image Builder archive file > >> names > >> + > >> endif > >> > >> > >> -- > >> 2.20.1 > >> > >> > >> _______________________________________________ > >> openwrt-devel mailing list > >> openwrt-devel@lists.openwrt.org > >> https://lists.openwrt.org/mailman/listinfo/openwrt-devel > > _______________________________________________ > openwrt-devel mailing list > openwrt-devel@lists.openwrt.org > https://lists.openwrt.org/mailman/listinfo/openwrt-devel
>>> I just think of ar71xx and ath79, where we have the same device but >> different targets. Of course, the name won't be exactly equal, as ath79 will >> have e.g. tplink_ prefix and ar71xx won't. >> Isn't ar71xx removed from master builds? It's neither at snapshot >> https://downloads.openwrt.org/snapshots/targets/ nor planed to be re- >> added to any upcoming release, is it? > > Yes, but it's just an example for a similar situation which might arise in the future. Then, we even might not have the current situation with the different device names, but may end up with completely identical names except for the target. > >>> For bcm63xx, we have two subtargets that build the same devices. >>> If we look at PR#2957, we might have a now bmips target at some point >> that features the same devices as bcm63xx. >> Can you please explain why that's the case? Why do we offer different >> images for the same device? I understand that for ar71xx -> ath79 within a > > I don't have any idea why this situation at bcm63xx exists; I just got aware of it at some point. Maybe Noltari or KanjiMonster can help ... > >> transfer period but it's never the scope to offer different "flavors" >> long term, is it? >>> This won't necessarily break anything, as images will still be in different >> folders (at least in /bin). >> I would be at least confusing and reverts the "unique profile name" idea. >>> However, we couldn't tell the difference between ar71xx/ath79 or similar >> from the image name (easily) after this change, or whether it's generic or >> smp for bcm63xx. For my personal taste, this drawback is bigger that the gain >> we will get from removing the target/subtarget part. >> Again, this sounds like a undesirable state where we not only build but also >> maintain multiple images for the save device. Wouldn't it be possible to add >> the target to all those "legacy images", however remove it wherever a target >> uses device tree and images.mk aka has long term support? > > Well, just look at the situation in 19.07. There we have both ar71xx and ath79 for the same devices, and even if we wanted, it would actually be quite hard to really filter out the ath79 devices in ar71xx. I really don't think removing the target from image names will pay out in the future. > >>> So, unless there is overwhelming support, I tend to NAK this. >> :( > > A compromise could be found by just removing the subtarget, but keeping the target in file names. This would mostly solve your problem with the generic names (at least there would be less duplicate info), but there would be significantly less situations where this was an impediment. Normally, no duplicate devices in a target exist, and if they are moved between subtargets, they are actually moved and not copied. The only remaining problem I can think of at the moment would be the bcm63xx situation, and maybe that one can be resolved at low cost. > > Best > > Adrian What about x86-{generic,legacy,64,...}? These subtargets each define a device just called "generic", with the image names only distinguished by their subtarget name. Matthias
On 6/15/20 7:32 AM, Matthias Schiffer wrote: >>>> I just think of ar71xx and ath79, where we have the same device but >>> different targets. Of course, the name won't be exactly equal, as ath79 will >>> have e.g. tplink_ prefix and ar71xx won't. >>> Isn't ar71xx removed from master builds? It's neither at snapshot >>> https://downloads.openwrt.org/snapshots/targets/ nor planed to be re- >>> added to any upcoming release, is it? >> Yes, but it's just an example for a similar situation which might arise in the future. Then, we even might not have the current situation with the different device names, but may end up with completely identical names except for the target. >> >>>> For bcm63xx, we have two subtargets that build the same devices. >>>> If we look at PR#2957, we might have a now bmips target at some point >>> that features the same devices as bcm63xx. >>> Can you please explain why that's the case? Why do we offer different >>> images for the same device? I understand that for ar71xx -> ath79 within a >> I don't have any idea why this situation at bcm63xx exists; I just got aware of it at some point. Maybe Noltari or KanjiMonster can help ... >> >>> transfer period but it's never the scope to offer different "flavors" >>> long term, is it? >>>> This won't necessarily break anything, as images will still be in different >>> folders (at least in /bin). >>> I would be at least confusing and reverts the "unique profile name" idea. >>>> However, we couldn't tell the difference between ar71xx/ath79 or similar >>> from the image name (easily) after this change, or whether it's generic or >>> smp for bcm63xx. For my personal taste, this drawback is bigger that the gain >>> we will get from removing the target/subtarget part. >>> Again, this sounds like a undesirable state where we not only build but also >>> maintain multiple images for the save device. Wouldn't it be possible to add >>> the target to all those "legacy images", however remove it wherever a target >>> uses device tree and images.mk aka has long term support? >> Well, just look at the situation in 19.07. There we have both ar71xx and ath79 for the same devices, and even if we wanted, it would actually be quite hard to really filter out the ath79 devices in ar71xx. I really don't think removing the target from image names will pay out in the future. >> >>>> So, unless there is overwhelming support, I tend to NAK this. >>> :( >> A compromise could be found by just removing the subtarget, but keeping the target in file names. This would mostly solve your problem with the generic names (at least there would be less duplicate info), but there would be significantly less situations where this was an impediment. Normally, no duplicate devices in a target exist, and if they are moved between subtargets, they are actually moved and not copied. The only remaining problem I can think of at the moment would be the bcm63xx situation, and maybe that one can be resolved at low cost. >> >> Best >> >> Adrian > > What about x86-{generic,legacy,64,...}? These subtargets each define a > device just called "generic", with the image names only distinguished by > their subtarget name. That's why I created this PR earlier, I should have created a patchset instead of splitting it on the ML and GitHub... https://github.com/openwrt/openwrt/pull/3082
diff --git a/include/image.mk b/include/image.mk index 984b64fb9c..c6fc467c9e 100644 --- a/include/image.mk +++ b/include/image.mk @@ -37,11 +37,12 @@ KDIR=$(KERNEL_BUILD_DIR) KDIR_TMP=$(KDIR)/tmp DTS_DIR:=$(LINUX_DIR)/arch/$(LINUX_KARCH)/boot/dts +IMG_PREFIX_TARGET:=$(if $(CONFIG_TARGET_FILENAMES),$(BOARD)$(if $(SUBTARGET),-$(SUBTARGET))-) IMG_PREFIX_EXTRA:=$(if $(EXTRA_IMAGE_NAME),$(call sanitize,$(EXTRA_IMAGE_NAME))-) IMG_PREFIX_VERNUM:=$(if $(CONFIG_VERSION_FILENAMES),$(call sanitize,$(VERSION_NUMBER))-) IMG_PREFIX_VERCODE:=$(if $(CONFIG_VERSION_CODE_FILENAMES),$(call sanitize,$(VERSION_CODE))-) -IMG_PREFIX:=$(VERSION_DIST_SANITIZED)-$(IMG_PREFIX_VERNUM)$(IMG_PREFIX_VERCODE)$(IMG_PREFIX_EXTRA)$(BOARD)$(if $(SUBTARGET),-$(SUBTARGET)) +IMG_PREFIX:=$(VERSION_DIST_SANITIZED)-$(IMG_PREFIX_VERNUM)$(IMG_PREFIX_VERCODE)$(IMG_PREFIX_EXTRA)$(IMG_PREFIX_TARGET) IMG_ROOTFS:=$(IMG_PREFIX)-rootfs IMG_COMBINED:=$(IMG_PREFIX)-combined IMG_PART_SIGNATURE:=$(shell echo $(SOURCE_DATE_EPOCH)$(LINUX_VERMAGIC) | mkhash md5 | cut -b1-8) @@ -293,7 +294,7 @@ endef define Image/Manifest $(call opkg,$(TARGET_DIR_ORIG)) list-installed > \ - $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),-$(PROFILE_SANITIZED)).manifest + $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),$(PROFILE_SANITIZED)).manifest endef define Image/gzip-ext4-padded-squashfs @@ -317,7 +318,7 @@ ifdef CONFIG_TARGET_ROOTFS_TARGZ define Image/Build/targz $(TAR) -cp --numeric-owner --owner=0 --group=0 --mode=a-s --sort=name \ $(if $(SOURCE_DATE_EPOCH),--mtime="@$(SOURCE_DATE_EPOCH)") \ - -C $(TARGET_DIR)/ . | gzip -9n > $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),-$(PROFILE_SANITIZED))-rootfs.tar.gz + -C $(TARGET_DIR)/ . | gzip -9n > $(BIN_DIR)/$(IMG_PREFIX)$(if $(PROFILE_SANITIZED),$(PROFILE_SANITIZED))-rootfs.tar.gz endef endif @@ -385,7 +386,7 @@ define Device/Init IMAGES := ARTIFACTS := - IMAGE_PREFIX := $(IMG_PREFIX)-$(1) + IMAGE_PREFIX := $(IMG_PREFIX)$(1) IMAGE_NAME = $$(IMAGE_PREFIX)-$$(1)-$$(2) IMAGE_SIZE := KERNEL_PREFIX = $$(IMAGE_PREFIX) diff --git a/package/base-files/image-config.in b/package/base-files/image-config.in index 3432db525a..5a70d51a7a 100644 --- a/package/base-files/image-config.in +++ b/package/base-files/image-config.in @@ -264,6 +264,15 @@ if VERSIONOPT Enable this to include the revision identifier or the configured version code into the firmware image, SDK- and Image Builder archive file names + + config TARGET_FILENAMES + bool + prompt "Target and subtarget in filenames" + default n + help + Enable this to include the target (and subtarget) in + firmware image, SDK- and Image Builder archive file names + endif
The target/subtarget information in image filenames is barely of any use for developers or end users. A developer reads the profile name and the target is either obvious due to previous work or `cd targets/ && grep -r <profile>` tells the target within 3ms. If no buildroot is available `cat <image> | tail -c 200` allows a look at the attached metadata which includes the target/subtarget. For users the information is entirely useless and maybe even harmful. Target names like `cortexa9` could easily be mistaken as an actual device name while the only relevant information would be `linksys_wrt3200acm`. Images are more realistically downloaded via a Wiki entry or a firmware wizard. This commit therefore adds the new image option called CONFIG_TARGET_FILENAMES to make the target/subtarget filename part optional. It is disabled by default. As the profile name `generic` appears multiple times in the x86 target as well as in oceton and ath25, the proposed patch on GitHub should be merged first: * treewide: use unique profile names #3082 https://github.com/openwrt/openwrt/pull/3082 Newly produced files would look like the following: * openwrt-linksys_wrt3200acm-initramfs-kernel.bin * openwrt-linksys_wrt3200acm.manifest * openwrt-linksys_wrt3200acm-squashfs-factory.img * openwrt-linksys_wrt3200acm-squashfs-sysupgrade.bin Signed-off-by: Paul Spooren <mail@aparcar.org> --- It's been a while since I made a controversial patch[0] so it feels about time. [0]: https://github.com/openwrt/openwrt/pull/2107 include/image.mk | 9 +++++---- package/base-files/image-config.in | 9 +++++++++ 2 files changed, 14 insertions(+), 4 deletions(-)