Kconfig: Print reverse dependencies in groups
v4 cover Kconfig: Print reverse dependencies in groups
From: Eugeniu Rosca <[email protected]>
Hello Masahiro, Ulf, Petr and all,
Here are a few words about the motivation behind this patch series.
First, the reason I got in touch with Kconfig is to optimize the configuration of automotive kernels, as well as to align the kernel configuration across a number of platforms. In the context of kernel optimization, one of the primary goals is to filter out any features that are not mentioned in the platform requirements.
Surprisingly or not, disabling a CONFIG option (which is assumed to be unneeded) may be not so trivial. Especially it is not trivial, when this CONFIG option is selected by a dozen of other configs. Before the moment commit 1ccb27143360 ("kconfig: make "Selected by:" and "Implied by:" readable") was submitted by Petr and eventually popped up in v4.16-rc1, it was an absolute pain to break down the "Selected by" reverse dependency expression in order to identify all those configs which select (IOW *do not allow disabling*) a certain feature (assumed to be not needed).
This patch series tries to make one step further and puts at users' fingertips the revdep top level OR sub-expressions grouped/clustered by the tristate value they evaluate to. This should allow the users to directly concentrate on and tackle the active reverse dependencies, which imho are the only ones that matter for a given ARCH and for a given defconfig (nevertheless we still print all of them).
Changes v3->v4 (fixed review findings from Ulf):
- Remove redundant default cases in switch constructs.
- Remove gettext _() tokens in str_append() calls.
- Aggregate code repetitions in expr_print_revdep().
Changes v2->v3:
- Switch from reverse dependencies prefixed by their tristate value to reverse dependencies grouped by the tristate value they evaluate to.
- Skip printing "{Selected,Implied} by [y|m|n]:" if there are no top level OR tokens/sub-expressions that evaluate to y|m|n (suggested by Petr).
- Use [1] as template for updating the interface/prototype of __expr_print() (suggested by Ulf).
Changes v1->v2:
- Don't skip the =n reverse dependency OR tokens, since some users might still need this information (suggested by Ulf).
- Instead of using "Selected by" for active tokens only, use it for all OR tokens, but specify the tristate value of each token as prefix (suggested by Masahiro).
[1] https://marc.info/?l=linux-kbuild&m=151777006005199&w=4
Eugeniu Rosca (3):
kconfig: Print reverse dependencies on new line consistently
kconfig: Prepare for printing reverse dependencies in groups
kconfig: Print reverse dependencies in groups
scripts/kconfig/expr.c | 102 +++++++++++++++++++++++++++++++++++++------- scripts/kconfig/expr.h | 11 ++++- scripts/kconfig/lkc_proto.h | 1 + scripts/kconfig/menu.c | 37 +++++++++++----- 4 files changed, 124 insertions(+), 27 deletions(-)
Signature · 1 line
2.16.1
PATCH v4 1/3 kconfig: Print reverse dependencies on new line consistently
From: Eugeniu Rosca <[email protected]>
Commit 1ccb27143360 ("kconfig: make "Selected by:" and "Implied by:" readable") made an incredible improvement in how reverse dependencies are perceived by the user, by breaking down the single (often interminable) expression string into small readable chunks, each of them displayed on a separate line:
Selected by:
- A && B
- C && (D || E)
Unfortunately, what happens with the non-OR (either E_SYMBOL or E_AND) expressions is that they don't get a dedicated line:
Selected by: F && G
That's arguably a bug/misbehavior, but it puts some amount of burden in implementing new ways of printing reverse dependencies to the user. As example, if we prefix every reverse dependency top level "||" token by its tristate value, then subjectively [2] looks more readable than [1].
[1] Selected by: [m] F && G
[2] Selected by:
- [m] F && G
Also, if we print the reverse dependency sub-expressions in groups (clustered by the tristate value they evaluate to), then subjectively [4] looks more readable than [3].
[3] Selected by [m]: F && G
[4] Selected by [m]:
- F && G
Based on the above, print all the reverse dependency sub-expressions on a separate line _consistently_. An example of change contributed by this commit is seen below.
| Symbol: NEED_SG_DMA_LENGTH [=y] | ... | Selected by: IOMMU_DMA [=y] && IOMMU_SUPPORT [=y]
becomes:
| Symbol: NEED_SG_DMA_LENGTH [=y] | ... | Selected by: | - IOMMU_DMA [=y] && IOMMU_SUPPORT [=y]
This patch has been tested using a tuned variant of zconfdump which prints the reverse dependencies for each config symbol.
--- scripts/kconfig/expr.c | 31 ++++++++++++++++++++++--------- 1 file changed, 22 insertions(+), 9 deletions(-)
scripts/kconfig/expr.c +22 -9
@@ -1179,6 +1179,16 @@ struct expr *expr_simplify_unmet_dep(struct expr *e1, struct expr *e2)return expr_get_leftmost_symbol (ret); } +static void +expr_print_newline (struct expr *e, +void (*fn)(void *,struct symbol *,const char *), +void *data, +int prevtoken) +{ +fn (data,NULL ,"\n - " ); +expr_print (e, fn, data, prevtoken); +} +static void __expr_print (struct expr *e,void (*fn)(void *,struct symbol *,const char *),void *data,int prevtoken,bool revdep) {if (!e) {
@@ -1191,7 +1201,10 @@ static void __expr_print(struct expr *e, void (*fn)(void *, struct symbol *, conswitch (e->type) {case E_SYMBOL :if (e->left.sym->name) -fn (data, e->left.sym, e->left.sym->name); +if (!revdep) +fn (data, e->left.sym, e->left.sym->name); +else +expr_print_newline (e, fn, data,E_OR );else fn (data,NULL ,"<choice>" );break ;
@@ -1234,19 +1247,19 @@ static void __expr_print(struct expr *e, void (*fn)(void *, struct symbol *, confn (data, e->right.sym, e->right.sym->name);break ;case E_OR : -if (revdep && e->left.expr->type !=E_OR ) -fn (data,NULL ,"\n - " );__expr_print (e->left.expr, fn, data,E_OR , revdep); -if (revdep) -fn (data,NULL ,"\n - " ); -else +if (!revdep)fn (data,NULL ," || " );__expr_print (e->right.expr, fn, data,E_OR , revdep);break ;case E_AND : -expr_print (e->left.expr, fn, data,E_AND ); -fn (data,NULL ," && " ); -expr_print (e->right.expr, fn, data,E_AND ); +if (!revdep) { +expr_print (e->left.expr, fn, data,E_AND ); +fn (data,NULL ," && " ); +expr_print (e->right.expr, fn, data,E_AND ); + }else { +expr_print_newline (e, fn, data,E_OR ); + }break ;case E_LIST :fn (data, e->right.sym, e->right.sym->name);
Signature · 1 line
2.16.1
PATCH v4 2/3 kconfig: Prepare for printing reverse dependencies in groups
In reply to Eugeniu RoscaFrom: Eugeniu Rosca <[email protected]>
Currently, reverse dependencies are printed like [1]. Prepare the ground for printing them like [2]. No functional change is intended in this patch.
[1] Current pattern for printing reverse dependencies:
Selected by:
- EXPR_A_Y /* evaluates to =y */
- EXPR_B_N /* evaluates to =n */
- EXPR_C_Y /* evaluates to =y */
- EXPR_D_M /* evaluates to =m */
- EXPR_E_N /* evaluates to =n */
[2] Upcoming pattern for printing reverse dependencies:
Selected by [y]:
- EXPR_A_Y
- EXPR_C_Y
Selected by [m]:
- EXPR_D_M
Selected by [n]:
- EXPR_B_N
- EXPR_E_N
--- scripts/kconfig/expr.c | 34 +++++++++++++++++++++++----------- scripts/kconfig/expr.h | 7 ++++++- scripts/kconfig/menu.c | 4 ++-- 3 files changed, 31 insertions(+), 14 deletions(-)
scripts/kconfig/expr.c +23 -11
@@ -1189,7 +1189,12 @@ expr_print_newline(struct expr *e,expr_print (e, fn, data, prevtoken); } -static void __expr_print (struct expr *e,void (*fn)(void *,struct symbol *,const char *),void *data,int prevtoken,bool revdep) +static void +__expr_print (struct expr *e, +void (*fn)(void *,struct symbol *,const char *), +void *data, +int prevtoken, +enum print_type type) {if (!e) {fn (data,NULL ,"y" );
@@ -1201,10 +1206,14 @@ static void __expr_print(struct expr *e, void (*fn)(void *, struct symbol *, conswitch (e->type) {case E_SYMBOL :if (e->left.sym->name) -if (!revdep) +switch (type) { +case PRINT_NORMAL :fn (data, e->left.sym, e->left.sym->name); -else +break ; +case PRINT_REVDEP_ALL :expr_print_newline (e, fn, data,E_OR ); +break ; + }else fn (data,NULL ,"<choice>" );break ;
@@ -1247,18 +1256,21 @@ static void __expr_print(struct expr *e, void (*fn)(void *, struct symbol *, confn (data, e->right.sym, e->right.sym->name);break ;case E_OR : -__expr_print (e->left.expr, fn, data,E_OR , revdep); -if (!revdep) +__expr_print (e->left.expr, fn, data,E_OR , type); +if (type ==PRINT_NORMAL )fn (data,NULL ," || " ); -__expr_print (e->right.expr, fn, data,E_OR , revdep); +__expr_print (e->right.expr, fn, data,E_OR , type);break ;case E_AND : -if (!revdep) { +switch (type) { +case PRINT_NORMAL :expr_print (e->left.expr, fn, data,E_AND );fn (data,NULL ," && " );expr_print (e->right.expr, fn, data,E_AND ); - }else { +break ; +case PRINT_REVDEP_ALL :expr_print_newline (e, fn, data,E_OR ); +break ; }break ;case E_LIST :
@@ -1289,7 +1301,7 @@ static void __expr_print(struct expr *e, void (*fn)(void *, struct symbol *, convoid expr_print (struct expr *e,void (*fn)(void *,struct symbol *,const char *),void *data,int prevtoken) { -__expr_print (e, fn, data, prevtoken, false); +__expr_print (e, fn, data, prevtoken,PRINT_NORMAL ); }static void expr_print_file_helper (void *data,struct symbol *sym,const char *str)
@@ -1342,7 +1354,7 @@ void expr_gstr_print(struct expr *e, struct gstr *gs)* line with a minus. This makes expressions much easier to read. * Suitable for reverse dependency expressions. */ -void expr_gstr_print_revdep (struct expr *e,struct gstr *gs) +void expr_gstr_print_revdep (struct expr *e,struct gstr *gs,enum print_type t) { -__expr_print (e, expr_print_gstr_helper, gs,E_NONE , true); +__expr_print (e, expr_print_gstr_helper, gs,E_NONE , t); }
scripts/kconfig/expr.h +6 -1
@@ -34,6 +34,11 @@ enum expr_type { E_LIST, E_SYMBOL, E_RANGE }; +enum print_type { +PRINT_NORMAL , +PRINT_REVDEP_ALL , +}; +union expr_data {struct expr *expr;struct symbol *sym;
@@ -310,7 +315,7 @@ struct expr *expr_simplify_unmet_dep(struct expr *e1, struct expr *e2);void expr_fprint (struct expr *e,FILE *out);struct gstr ;/* forward */ void expr_gstr_print (struct expr *e,struct gstr *gs); -void expr_gstr_print_revdep (struct expr *e,struct gstr *gs); +void expr_gstr_print_revdep (struct expr *e,struct gstr *gs,enum print_type t);static inline int expr_is_yes (struct expr *e) {
scripts/kconfig/menu.c +2 -2
@@ -828,14 +828,14 @@ static void get_symbol_str(struct gstr *r, struct symbol *sym,get_symbol_props_str (r, sym,P_SELECT ,_ (" Selects: " ));if (sym->rev_dep.expr) {str_append (r,_ (" Selected by: " )); -expr_gstr_print_revdep (sym->rev_dep.expr, r); +expr_gstr_print_revdep (sym->rev_dep.expr, r,PRINT_REVDEP_ALL );str_append (r,"\n" ); }get_symbol_props_str (r, sym,P_IMPLY ,_ (" Implies: " ));if (sym->implied.expr) {str_append (r,_ (" Implied by: " )); -expr_gstr_print_revdep (sym->implied.expr, r); +expr_gstr_print_revdep (sym->implied.expr, r,PRINT_REVDEP_ALL );str_append (r,"\n" ); }
Signature · 1 line
2.16.1
PATCH v4 3/3 kconfig: Print reverse dependencies in groups
In reply to Eugeniu RoscaFrom: Eugeniu Rosca <[email protected]>
Assuming commit 617aebe6a97e ("Merge tag 'usercopy-v4.16-rc1' ofgit://git.kernel.org/pub/scm/linux/kernel/git/kees/linux"), ARCH=arm64 and vanilla arm64 defconfig, here is the top 10 CONFIG options with the highest amount of top level "||" sub-expressions/tokens that make up the final "{Selected,Implied} by" reverse dependency expression.
| Config | Revdep all | Revdep ![=n] | |-------------------------------|------------|--------------| | REGMAP_I2C | 212 | 9 | | CRC32 | 167 | 25 | | FW_LOADER | 128 | 5 | | MFD_CORE | 124 | 9 | | FB_CFB_IMAGEBLIT | 114 | 2 | | FB_CFB_COPYAREA | 111 | 2 | | FB_CFB_FILLRECT | 110 | 2 | | SND_PCM | 103 | 2 | | CRYPTO_HASH | 87 | 19 | | WATCHDOG_CORE | 86 | 6 |
The story behind the above is that users need to visually review/evaluate 212 expressions which *potentially* select REGMAP_I2C in order to identify the expressions which *actually* select REGMAP_I2C, for a particular ARCH and for a particular defconfig used.
To make this experience smoother, change the way reverse dependencies are displayed to the user from [1] to [2].
[1] Old representation of reverse dependencies for DMA_ENGINE_RAID:
Selected by:
- AMCC_PPC440SPE_ADMA [=n] && DMADEVICES [=y] && (440SPe || 440SP)
- BCM_SBA_RAID [=m] && DMADEVICES [=y] && (ARM64 [=y] || ...
- FSL_RAID [=n] && DMADEVICES [=y] && FSL_SOC && ...
- INTEL_IOATDMA [=n] && DMADEVICES [=y] && PCI [=y] && X86_64
- MV_XOR [=n] && DMADEVICES [=y] && (PLAT_ORION || ARCH_MVEBU [=y] ...
- MV_XOR_V2 [=y] && DMADEVICES [=y] && ARM64 [=y]
- XGENE_DMA [=n] && DMADEVICES [=y] && (ARCH_XGENE [=y] || ...
- DMATEST [=n] && DMADEVICES [=y] && DMA_ENGINE [=y]
[2] New representation of reverse dependencies for DMA_ENGINE_RAID:
Selected by [y]:
- MV_XOR_V2 [=y] && DMADEVICES [=y] && ARM64 [=y]
Selected by [m]:
- BCM_SBA_RAID [=m] && DMADEVICES [=y] && (ARM64 [=y] || ... Selected by [n]:
- AMCC_PPC440SPE_ADMA [=n] && DMADEVICES [=y] && (440SPe || ...
- FSL_RAID [=n] && DMADEVICES [=y] && FSL_SOC && ...
- INTEL_IOATDMA [=n] && DMADEVICES [=y] && PCI [=y] && X86_64
- MV_XOR [=n] && DMADEVICES [=y] && (PLAT_ORION || ARCH_MVEBU [=y] ...
- XGENE_DMA [=n] && DMADEVICES [=y] && (ARCH_XGENE [=y] || ...
- DMATEST [=n] && DMADEVICES [=y] && DMA_ENGINE [=y]
--- scripts/kconfig/expr.c | 63 ++++++++++++++++++++++++++++++++++++++------- scripts/kconfig/expr.h | 4 +++ scripts/kconfig/lkc_proto.h | 1 + scripts/kconfig/menu.c | 37 +++++++++++++++++++------- 4 files changed, 86 insertions(+), 19 deletions(-)
scripts/kconfig/expr.c +54 -9
@@ -1180,13 +1180,19 @@ struct expr *expr_simplify_unmet_dep(struct expr *e1, struct expr *e2) }static void -expr_print_newline (struct expr *e, -void (*fn)(void *,struct symbol *,const char *), -void *data, -int prevtoken) +expr_print_revdep (struct expr *e, +void (*fn)(void *,struct symbol *,const char *), +void *data, +int prevtoken, +enum print_type type) { -fn (data,NULL ,"\n - " ); -expr_print (e, fn, data, prevtoken); +if (type ==PRINT_REVDEP_ALL || + type ==PRINT_REVDEP_YES &&expr_calc_value (e) == yes || + type ==PRINT_REVDEP_MOD &&expr_calc_value (e) == mod || + type ==PRINT_REVDEP_NO &&expr_calc_value (e) == no) { +fn (data,NULL ,"\n - " ); +expr_print (e, fn, data, prevtoken); + } } static void
@@ -1211,7 +1217,10 @@ __expr_print(struct expr *e,fn (data, e->left.sym, e->left.sym->name);break ;case PRINT_REVDEP_ALL : -expr_print_newline (e, fn, data,E_OR ); +case PRINT_REVDEP_YES : +case PRINT_REVDEP_MOD : +case PRINT_REVDEP_NO : +expr_print_revdep (e, fn, data,E_OR , type);break ; }else
@@ -1269,7 +1278,10 @@ __expr_print(struct expr *e,expr_print (e->right.expr, fn, data,E_AND );break ;case PRINT_REVDEP_ALL : -expr_print_newline (e, fn, data,E_OR ); +case PRINT_REVDEP_YES : +case PRINT_REVDEP_MOD : +case PRINT_REVDEP_NO : +expr_print_revdep (e, fn, data,E_OR , type);break ; }break ;
@@ -1349,10 +1361,43 @@ void expr_gstr_print(struct expr *e, struct gstr *gs)expr_print (e, expr_print_gstr_helper, gs,E_NONE ); } +/* +* Allow front ends to check if a specific reverse dependency expression +* has at least one top level "||" member which evaluates to "val". This, +* will allow front ends to, as example, avoid printing "Selected by [y]:" +* line when there are actually no top level "||" sub-expressions which +* evaluate to =y. +*/ +bool expr_revdep_contains (struct expr *e,tristate val) +{ +bool ret = false; + +if (!e) +return ret; + +switch (e->type) { +case E_SYMBOL : +case E_AND : +if (expr_calc_value (e) == val) + ret = true; +break ; +case E_OR : +if (expr_revdep_contains (e->left.expr, val)) + ret = true; +else if (expr_revdep_contains (e->right.expr, val)) + ret = true; +break ; +default : +break ; + } +return ret; +} +/* * Transform the top level "||" tokens into newlines and prepend each * line with a minus. This makes expressions much easier to read. -* Suitable for reverse dependency expressions. +* Suitable for reverse dependency expressions. In addition, allow +* selective printing of tokens/sub-expressions by their tristate value. */ void expr_gstr_print_revdep (struct expr *e,struct gstr *gs,enum print_type t) {
scripts/kconfig/expr.h +4 -0
@@ -37,6 +37,9 @@ enum expr_type {enum print_type {PRINT_NORMAL ,PRINT_REVDEP_ALL , +PRINT_REVDEP_YES , +PRINT_REVDEP_MOD , +PRINT_REVDEP_NO , };union expr_data {
@@ -316,6 +319,7 @@ void expr_fprint(struct expr *e, FILE *out);struct gstr ;/* forward */ void expr_gstr_print (struct expr *e,struct gstr *gs);void expr_gstr_print_revdep (struct expr *e,struct gstr *gs,enum print_type t); +bool expr_revdep_contains (struct expr *e,tristate val);static inline int expr_is_yes (struct expr *e) {
scripts/kconfig/lkc_proto.h +1 -0
@@ -25,6 +25,7 @@ bool menu_has_help(struct menu *menu);const char *menu_get_help (struct menu *menu);struct gstr get_relations_str (struct symbol **sym_arr,struct list_head *head);void menu_get_ext_help (struct menu *menu,struct gstr *help); +void get_revdep_by_type (struct expr *e,char *s,struct gstr *r);/* symbol.c */ extern struct symbol * symbol_hash[SYMBOL_HASHSIZE ];
scripts/kconfig/menu.c +27 -10
@@ -790,6 +790,31 @@ static void get_symbol_props_str(struct gstr *r, struct symbol *sym,str_append (r,"\n" ); } +void get_revdep_by_type (struct expr *e,char *s,struct gstr *r) +{ +if (!e) +return ; + +if (expr_revdep_contains (e, yes)) { +str_append (r, s); +str_append (r," [y]:" ); +expr_gstr_print_revdep (e, r,PRINT_REVDEP_YES ); +str_append (r,"\n" ); + } +if (expr_revdep_contains (e, mod)) { +str_append (r, s); +str_append (r," [m]:" ); +expr_gstr_print_revdep (e, r,PRINT_REVDEP_MOD ); +str_append (r,"\n" ); + } +if (expr_revdep_contains (e, no)) { +str_append (r, s); +str_append (r," [n]:" ); +expr_gstr_print_revdep (e, r,PRINT_REVDEP_NO ); +str_append (r,"\n" ); + } +} +/* * head is optional and may be NULL */
@@ -826,18 +851,10 @@ static void get_symbol_str(struct gstr *r, struct symbol *sym, }get_symbol_props_str (r, sym,P_SELECT ,_ (" Selects: " )); -if (sym->rev_dep.expr) { -str_append (r,_ (" Selected by: " )); -expr_gstr_print_revdep (sym->rev_dep.expr, r,PRINT_REVDEP_ALL ); -str_append (r,"\n" ); - } +get_revdep_by_type (sym->rev_dep.expr," Selected by" , r);get_symbol_props_str (r, sym,P_IMPLY ,_ (" Implies: " )); -if (sym->implied.expr) { -str_append (r,_ (" Implied by: " )); -expr_gstr_print_revdep (sym->implied.expr, r,PRINT_REVDEP_ALL ); -str_append (r,"\n" ); - } +get_revdep_by_type (sym->implied.expr," Implied by" , r);str_append (r,"\n\n" ); }
Signature · 1 line
2.16.1
Quoted 57 lines · Eugeniu Rosca
From: Eugeniu Rosca <[email protected]>
Hello Masahiro, Ulf, Petr and all,
Here are a few words about the motivation behind this patch series.
First, the reason I got in touch with
Kconfigis to optimize the configuration of automotive kernels, as well as to align the kernel configuration across a number of platforms. In the context of kernel optimization, one of the primary goals is to filter out any features that are not mentioned in the platform requirements.Surprisingly or not, disabling a CONFIG option (which is assumed to be unneeded) may be not so trivial. Especially it is not trivial, when this CONFIG option is selected by a dozen of other configs. Before the moment commit
1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable") was submitted by Petr and eventually popped up in v4.16-rc1, it was an absolute pain to break down the "Selected by" reverse dependency expression in order to identify all those configs which select (IOW *do not allow disabling*) a certain feature (assumed to be not needed).This patch series tries to make one step further and puts at users' fingertips the revdep top level OR sub-expressions grouped/clustered by the tristate value they evaluate to. This should allow the users to directly concentrate on and tackle the active reverse dependencies, which imho are the only ones that matter for a given ARCH and for a given defconfig (nevertheless we still print all of them).
Changes
v3->v4(fixed review findings from Ulf):
- Remove redundant default cases in switch constructs.
- Remove gettext_()tokens instr_append()calls.
- Aggregate code repetitions inexpr_print_revdep().Changes
v2->v3:
- Switch from reverse dependencies prefixed by their tristate value to reverse dependencies grouped by the tristate value they evaluate to.
- Skip printing "{Selected,Implied} by [y|m|n]:" if there are no top level OR tokens/sub-expressions that evaluate to y|m|n (suggested by Petr).
- Use [1] as template for updating the interface/prototype of__expr_print()(suggested by Ulf).Changes
v1->v2:
- Don't skip the =n reverse dependency OR tokens, since some users might still need this information (suggested by Ulf).
- Instead of using "Selected by" for active tokens only, use it for all OR tokens, but specify the tristate value of each token as prefix (suggested by Masahiro).[1] https://marc.info/?l=linux-kbuild&m=151777006005199&w=4
Eugeniu Rosca (3):
kconfig: Print reverse dependencies on new line consistently
kconfig: Prepare for printing reverse dependencies in groups
kconfig: Print reverse dependencies in groups
Quoted 9 lines
scripts/kconfig/expr.c | 102 +++++++++++++++++++++++++++++++++++++------- scripts/kconfig/expr.h | 11 ++++- scripts/kconfig/lkc_proto.h | 1 + scripts/kconfig/menu.c | 37 +++++++++++----- 4 files changed, 124 insertions(+), 27 deletions(-)--
2.16.1
Whole patchset looks fine to me now:
Cheers,
Ulf
2018-02-19 5:47 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 57 lines
From: Eugeniu Rosca <[email protected]>
Hello Masahiro, Ulf, Petr and all,
Here are a few words about the motivation behind this patch series.
First, the reason I got in touch with
Kconfigis to optimize the configuration of automotive kernels, as well as to align the kernel configuration across a number of platforms. In the context of kernel optimization, one of the primary goals is to filter out any features that are not mentioned in the platform requirements.Surprisingly or not, disabling a CONFIG option (which is assumed to be unneeded) may be not so trivial. Especially it is not trivial, when this CONFIG option is selected by a dozen of other configs. Before the moment commit
1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable") was submitted by Petr and eventually popped up in v4.16-rc1, it was an absolute pain to break down the "Selected by" reverse dependency expression in order to identify all those configs which select (IOW *do not allow disabling*) a certain feature (assumed to be not needed).This patch series tries to make one step further and puts at users' fingertips the revdep top level OR sub-expressions grouped/clustered by the tristate value they evaluate to. This should allow the users to directly concentrate on and tackle the active reverse dependencies, which imho are the only ones that matter for a given ARCH and for a given defconfig (nevertheless we still print all of them).
Changes
v3->v4(fixed review findings from Ulf):
- Remove redundant default cases in switch constructs.
- Remove gettext_()tokens instr_append()calls.
- Aggregate code repetitions inexpr_print_revdep().Changes
v2->v3:
- Switch from reverse dependencies prefixed by their tristate value to reverse dependencies grouped by the tristate value they evaluate to.
- Skip printing "{Selected,Implied} by [y|m|n]:" if there are no top level OR tokens/sub-expressions that evaluate to y|m|n (suggested by Petr).
- Use [1] as template for updating the interface/prototype of__expr_print()(suggested by Ulf).Changes
v1->v2:
- Don't skip the =n reverse dependency OR tokens, since some users might still need this information (suggested by Ulf).
- Instead of using "Selected by" for active tokens only, use it for all OR tokens, but specify the tristate value of each token as prefix (suggested by Masahiro).[1] https://marc.info/?l=linux-kbuild&m=151777006005199&w=4
Eugeniu Rosca (3):
kconfig: Print reverse dependencies on new line consistently
kconfig: Prepare for printing reverse dependencies in groups
kconfig: Print reverse dependencies in groups
Quoted 6 lines
scripts/kconfig/expr.c | 102 +++++++++++++++++++++++++++++++++++++------- scripts/kconfig/expr.h | 11 ++++- scripts/kconfig/lkc_proto.h | 1 + scripts/kconfig/menu.c | 37 +++++++++++----- 4 files changed, 124 insertions(+), 27 deletions(-)
I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.
Please rewrite the code within 20 lines.
Signature · 2 lines
Best Regards Masahiro Yamada
2018-02-20 17:13 GMT+09:00 Masahiro Yamada <[email protected]>:
Quoted 66 lines
2018-02-19 5:47 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 57 lines
From: Eugeniu Rosca <[email protected]>
Hello Masahiro, Ulf, Petr and all,
Here are a few words about the motivation behind this patch series.
First, the reason I got in touch with
Kconfigis to optimize the configuration of automotive kernels, as well as to align the kernel configuration across a number of platforms. In the context of kernel optimization, one of the primary goals is to filter out any features that are not mentioned in the platform requirements.Surprisingly or not, disabling a CONFIG option (which is assumed to be unneeded) may be not so trivial. Especially it is not trivial, when this CONFIG option is selected by a dozen of other configs. Before the moment commit
1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable") was submitted by Petr and eventually popped up in v4.16-rc1, it was an absolute pain to break down the "Selected by" reverse dependency expression in order to identify all those configs which select (IOW *do not allow disabling*) a certain feature (assumed to be not needed).This patch series tries to make one step further and puts at users' fingertips the revdep top level OR sub-expressions grouped/clustered by the tristate value they evaluate to. This should allow the users to directly concentrate on and tackle the active reverse dependencies, which imho are the only ones that matter for a given ARCH and for a given defconfig (nevertheless we still print all of them).
Changes
v3->v4(fixed review findings from Ulf):
- Remove redundant default cases in switch constructs.
- Remove gettext_()tokens instr_append()calls.
- Aggregate code repetitions inexpr_print_revdep().Changes
v2->v3:
- Switch from reverse dependencies prefixed by their tristate value to reverse dependencies grouped by the tristate value they evaluate to.
- Skip printing "{Selected,Implied} by [y|m|n]:" if there are no top level OR tokens/sub-expressions that evaluate to y|m|n (suggested by Petr).
- Use [1] as template for updating the interface/prototype of__expr_print()(suggested by Ulf).Changes
v1->v2:
- Don't skip the =n reverse dependency OR tokens, since some users might still need this information (suggested by Ulf).
- Instead of using "Selected by" for active tokens only, use it for all OR tokens, but specify the tristate value of each token as prefix (suggested by Masahiro).[1] https://marc.info/?l=linux-kbuild&m=151777006005199&w=4
Eugeniu Rosca (3):
kconfig: Print reverse dependencies on new line consistently
kconfig: Prepare for printing reverse dependencies in groups
kconfig: Print reverse dependencies in groupsQuoted 6 lines
scripts/kconfig/expr.c | 102 +++++++++++++++++++++++++++++++++++++------- scripts/kconfig/expr.h | 11 ++++- scripts/kconfig/lkc_proto.h | 1 + scripts/kconfig/menu.c | 37 +++++++++++----- 4 files changed, 124 insertions(+), 27 deletions(-)
Quoted 5 lines
I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.Please rewrite the code within 20 lines.
For a hint, I cleaned up the code base.
https://patchwork.kernel.org/patch/10229545/
which should be equivalent to yours:
https://patchwork.kernel.org/patch/10226951/
No 'enum print_type', please.
Signature · 2 lines
Best Regards Masahiro Yamada
Quoted 73 lines · Masahiro Yamada
2018-02-20 17:13 GMT+09:00 Masahiro Yamada <[email protected]>:
Quoted 66 lines
2018-02-19 5:47 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 57 lines
From: Eugeniu Rosca <[email protected]>
Hello Masahiro, Ulf, Petr and all,
Here are a few words about the motivation behind this patch series.
First, the reason I got in touch with
Kconfigis to optimize the configuration of automotive kernels, as well as to align the kernel configuration across a number of platforms. In the context of kernel optimization, one of the primary goals is to filter out any features that are not mentioned in the platform requirements.Surprisingly or not, disabling a CONFIG option (which is assumed to be unneeded) may be not so trivial. Especially it is not trivial, when this CONFIG option is selected by a dozen of other configs. Before the moment commit
1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable") was submitted by Petr and eventually popped up in v4.16-rc1, it was an absolute pain to break down the "Selected by" reverse dependency expression in order to identify all those configs which select (IOW *do not allow disabling*) a certain feature (assumed to be not needed).This patch series tries to make one step further and puts at users' fingertips the revdep top level OR sub-expressions grouped/clustered by the tristate value they evaluate to. This should allow the users to directly concentrate on and tackle the active reverse dependencies, which imho are the only ones that matter for a given ARCH and for a given defconfig (nevertheless we still print all of them).
Changes
v3->v4(fixed review findings from Ulf):
- Remove redundant default cases in switch constructs.
- Remove gettext_()tokens instr_append()calls.
- Aggregate code repetitions inexpr_print_revdep().Changes
v2->v3:
- Switch from reverse dependencies prefixed by their tristate value to reverse dependencies grouped by the tristate value they evaluate to.
- Skip printing "{Selected,Implied} by [y|m|n]:" if there are no top level OR tokens/sub-expressions that evaluate to y|m|n (suggested by Petr).
- Use [1] as template for updating the interface/prototype of__expr_print()(suggested by Ulf).Changes
v1->v2:
- Don't skip the =n reverse dependency OR tokens, since some users might still need this information (suggested by Ulf).
- Instead of using "Selected by" for active tokens only, use it for all OR tokens, but specify the tristate value of each token as prefix (suggested by Masahiro).[1] https://marc.info/?l=linux-kbuild&m=151777006005199&w=4
Eugeniu Rosca (3):
kconfig: Print reverse dependencies on new line consistently
kconfig: Prepare for printing reverse dependencies in groups
kconfig: Print reverse dependencies in groupsQuoted 6 lines
scripts/kconfig/expr.c | 102 +++++++++++++++++++++++++++++++++++++------- scripts/kconfig/expr.h | 11 ++++- scripts/kconfig/lkc_proto.h | 1 + scripts/kconfig/menu.c | 37 +++++++++++----- 4 files changed, 124 insertions(+), 27 deletions(-)Quoted 5 lines
I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.Please rewrite the code within 20 lines.
Quoted 8 lines
For a hint, I cleaned up the code base.
https://patchwork.kernel.org/patch/10229545/which should be equivalent to yours:
https://patchwork.kernel.org/patch/10226951/No 'enum
print_type', please.
The reason I prefer them on separate lines consistently is to avoid stuff like the following:
Selected by [y]: MV_XOR_V2 [=y] && DMADEVICES [=y] && ARM64 [=y] Selected by [n]:
- AMCC_PPC440SPE_ADMA [=n] && DMADEVICES [=y] && (440SPe || ...
- FSL_RAID [=n] && DMADEVICES [=y] && FSL_SOC && ...
- INTEL_IOATDMA [=n] && DMADEVICES [=y] && PCI [=y] && X86_64
That looks confusing and unbalanced to me.
There are some simple ways to trim down the size of this patchset though.
Eugeniu:
What do you think about the following refactoring of your 3/3 patch? Maybe there's a way to have expr_print_revdep() take just a tristate too, though it's not worth it if it just grows the code elsewhere.
(By the way, I noticed that expr_print_revdep() previously generated a warning suggesting parentheses, which was my fault. If you saw that, don't copy my mistakes. The build should be warning-free. :)
scripts/kconfig/expr.c +13 -40
@@ -1186,10 +1186,9 @@ expr_print_revdep(struct expr *e, int prevtoken, enum print_type type) { -if (type ==PRINT_REVDEP_ALL || - type ==PRINT_REVDEP_YES &&expr_calc_value (e) == yes || - type ==PRINT_REVDEP_MOD &&expr_calc_value (e) == mod || - type ==PRINT_REVDEP_NO &&expr_calc_value (e) == no) { +if ((type ==PRINT_REVDEP_YES &&expr_calc_value (e) == yes) || + (type ==PRINT_REVDEP_MOD &&expr_calc_value (e) == mod) || + (type ==PRINT_REVDEP_NO &&expr_calc_value (e) == no)) {fn (data,NULL ,"\n - " );expr_print (e, fn, data, prevtoken); }
@@ -1212,17 +1211,10 @@ __expr_print(struct expr *e,switch (e->type) {case E_SYMBOL :if (e->left.sym->name) -switch (type) { -case PRINT_NORMAL : +if (type ==PRINT_NORMAL )fn (data, e->left.sym, e->left.sym->name); -break ; -case PRINT_REVDEP_ALL : -case PRINT_REVDEP_YES : -case PRINT_REVDEP_MOD : -case PRINT_REVDEP_NO : +else expr_print_revdep (e, fn, data,E_OR , type); -break ; - }else fn (data,NULL ,"<choice>" );break ;
@@ -1271,18 +1263,12 @@ __expr_print(struct expr *e,__expr_print (e->right.expr, fn, data,E_OR , type);break ;case E_AND : -switch (type) { -case PRINT_NORMAL : +if (type ==PRINT_NORMAL ) {expr_print (e->left.expr, fn, data,E_AND );fn (data,NULL ," && " );expr_print (e->right.expr, fn, data,E_AND ); -break ; -case PRINT_REVDEP_ALL : -case PRINT_REVDEP_YES : -case PRINT_REVDEP_MOD : -case PRINT_REVDEP_NO : + }else {expr_print_revdep (e, fn, data,E_OR , type); -break ; }break ;case E_LIST :
@@ -1370,27 +1356,14 @@ void expr_gstr_print(struct expr *e, struct gstr *gs)*/ bool expr_revdep_contains (struct expr *e,tristate val) { -bool ret = false; -if (!e) -return ret; +return false; -switch (e->type) { -case E_SYMBOL : -case E_AND : -if (expr_calc_value (e) == val) - ret = true; -break ; -case E_OR : -if (expr_revdep_contains (e->left.expr, val)) - ret = true; -else if (expr_revdep_contains (e->right.expr, val)) - ret = true; -break ; -default : -break ; - } -return ret; +if (e->type ==E_OR ) +return expr_revdep_contains (e->left.expr, val) || +expr_revdep_contains (e->right.expr, val); + +return expr_calc_value (e) == val; }/*
scripts/kconfig/expr.h +0 -1
@@ -36,7 +36,6 @@ enum expr_type {enum print_type {PRINT_NORMAL , -PRINT_REVDEP_ALL ,PRINT_REVDEP_YES ,PRINT_REVDEP_MOD ,PRINT_REVDEP_NO ,
Quoted 20 lines · Ulf Magnusson
Quoted 8 lines · Masahiro Yamada
2018-02-20 17:13 GMT+09:00 Masahiro Yamada <[email protected]>:
Quoted 6 lines
I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.Please rewrite the code within 20 lines.
Quoted 8 lines
For a hint, I cleaned up the code base.
https://patchwork.kernel.org/patch/10229545/which should be equivalent to yours:
https://patchwork.kernel.org/patch/10226951/No 'enum
print_type', please.The reason I prefer them on separate lines consistently is to avoid stuff like the following:
Quoted 7 lines
Selected by [y]:
MV_XOR_V2[=y] && DMADEVICES [=y] && ARM64 [=y] Selected by [n]:
-AMCC_PPC440SPE_ADMA[=n] && DMADEVICES [=y] && (440SPe || ...
-FSL_RAID[=n] && DMADEVICES [=y] &&FSL_SOC&& ...
-INTEL_IOATDMA[=n] && DMADEVICES [=y] && PCI [=y] &&X86_64That looks confusing and unbalanced to me.
I think nobody is disputing this. Masahiro's
https://patchwork.kernel.org/patch/10229545/ is also showing every reverse dependency top level OR expression on a new row.
Quoted 9 lines
There are some simple ways to trim down the size of this patchset though.
Eugeniu:
What do you think about the following refactoring of your 3/3 patch? Maybe there's a way to haveexpr_print_revdep()take just a tristate too, though it's not worth it if it just grows the code elsewhere.
Well, Masahiro requests not using 'enum print_type' and I still see it in your example.
Quoted 3 lines
(By the way, I noticed that
expr_print_revdep()previously generated a warning suggesting parentheses, which was my fault. If you saw that, don't copy my mistakes. The build should be warning-free. :)
I didn't notice any warnings using gcc 5.4.0.
Quoted 82 lines
scripts/kconfig/expr.c +8
-36@@ -1186,10 +1186,9 @@ expr_print_revdep(struct expr *e, int prevtoken, enum print_type type) { -if (type ==PRINT_REVDEP_ALL || - type ==PRINT_REVDEP_YES &&expr_calc_value (e) == yes || - type ==PRINT_REVDEP_MOD &&expr_calc_value (e) == mod || - type ==PRINT_REVDEP_NO &&expr_calc_value (e) == no) { +if ((type ==PRINT_REVDEP_YES &&expr_calc_value (e) == yes) || + (type ==PRINT_REVDEP_MOD &&expr_calc_value (e) == mod) || + (type ==PRINT_REVDEP_NO &&expr_calc_value (e) == no)) {fn (data,NULL ,"\n - " );expr_print (e, fn, data, prevtoken); }@@ -1212,17 +1211,10 @@ __expr_print(struct expr *e,switch (e->type) {case E_SYMBOL :if (e->left.sym->name) -switch (type) { -case PRINT_NORMAL : +if (type ==PRINT_NORMAL )fn (data, e->left.sym, e->left.sym->name); -break ; -case PRINT_REVDEP_ALL : -case PRINT_REVDEP_YES : -case PRINT_REVDEP_MOD : -case PRINT_REVDEP_NO : +else expr_print_revdep (e, fn, data,E_OR , type); -break ; - }else fn (data,NULL ,"<choice>" );break ;@@ -1271,18 +1263,12 @@ __expr_print(struct expr *e,__expr_print (e->right.expr, fn, data,E_OR , type);break ;case E_AND : -switch (type) { -case PRINT_NORMAL : +if (type ==PRINT_NORMAL ) {expr_print (e->left.expr, fn, data,E_AND );fn (data,NULL ," && " );expr_print (e->right.expr, fn, data,E_AND ); -break ; -case PRINT_REVDEP_ALL : -case PRINT_REVDEP_YES : -case PRINT_REVDEP_MOD : -case PRINT_REVDEP_NO : + }else {expr_print_revdep (e, fn, data,E_OR , type); -break ; }break ;case E_LIST :@@ -1370,27 +1356,14 @@ void expr_gstr_print(struct expr *e, struct gstr *gs)*/ bool expr_revdep_contains (struct expr *e,tristate val) { -bool ret = false; -if (!e) -return ret; +return false; -switch (e->type) { -case E_SYMBOL : -case E_AND : -if (expr_calc_value (e) == val) - ret = true; -break ; -case E_OR : -if (expr_revdep_contains (e->left.expr, val)) - ret = true; -else if (expr_revdep_contains (e->right.expr, val)) - ret = true; -break ;
scripts/kconfig/expr.c void expr_gstr_print(struct expr *e, struct gstr *gs) · line 1367
Quoted 10 lines
scripts/kconfig/expr.c +5
-4-default : -break ; - } -return ret; +if (e->type ==E_OR ) +return expr_revdep_contains (e->left.expr, val) || +expr_revdep_contains (e->right.expr, val); + +return expr_calc_value (e) == val; }
I really like how you've minimized the size of expr_revdep_contains(). Thanks for that. I will try to come up with something in the next days.
Quoted 7 lines
scripts/kconfig/expr.c +0
-0/* scripts/kconfig/expr.h +0
-0
scripts/kconfig/expr.h enum expr_type { · line 41
Quoted 7 lines
scripts/kconfig/expr.h +0
-1@@ -36,7 +36,6 @@ enum expr_type {enum print_type {PRINT_NORMAL , -PRINT_REVDEP_ALL ,PRINT_REVDEP_YES ,PRINT_REVDEP_MOD ,PRINT_REVDEP_NO ,
Best regards,
Eugeniu.
Quoted 27 lines · Eugeniu Rosca
Quoted 26 lines · Ulf Magnusson
Quoted 8 lines · Masahiro Yamada
2018-02-20 17:13 GMT+09:00 Masahiro Yamada <[email protected]>:
Quoted 6 lines
I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.Please rewrite the code within 20 lines.
Quoted 8 lines
For a hint, I cleaned up the code base.
https://patchwork.kernel.org/patch/10229545/which should be equivalent to yours:
https://patchwork.kernel.org/patch/10226951/No 'enum
print_type', please.The reason I prefer them on separate lines consistently is to avoid stuff like the following:
Selected by [y]:
MV_XOR_V2[=y] && DMADEVICES [=y] && ARM64 [=y] Selected by [n]:
-AMCC_PPC440SPE_ADMA[=n] && DMADEVICES [=y] && (440SPe || ...
-FSL_RAID[=n] && DMADEVICES [=y] &&FSL_SOC&& ...
-INTEL_IOATDMA[=n] && DMADEVICES [=y] && PCI [=y] &&X86_64
Quoted 5 lines
Quoted 1 line
That looks confusing and unbalanced to me.
I think nobody is disputing this. Masahiro's
https://patchwork.kernel.org/patch/10229545/ is also showing every reverse dependency top level OR expression on a new row.
Ohh... sorry, didn't read closely enough.
Yeah, I agree that that looks like a nice and neat solution.
Quoted 5 lines
Quoted 5 lines
There are some simple ways to trim down the size of this patchset though.
Quoted 8 lines
Quoted 5 lines
Eugeniu:
What do you think about the following refactoring of your 3/3 patch? Maybe there's a way to haveexpr_print_revdep()take just a tristate too, though it's not worth it if it just grows the code elsewhere.Well, Masahiro requests not using 'enum
print_type' and I still see it in your example.
One alternative would be to keep the 'revdep' flag and add an extra tristate parameter for the kinds of selects to print.
I'm not a huge fan of having parameters that turn meaningless depending on the values of other parameters, but it might not be terrible there, if we must get rid of 'enum print_type'. I'm always open to other suggestions too.
The whole expression printing code feels a bit twisty to me with all the mutual recursion and stuff going on, but that's outside the scope of this patchset. :)
Quoted 6 lines
Quoted 4 lines
(By the way, I noticed that
expr_print_revdep()previously generated a warning suggesting parentheses, which was my fault. If you saw that, don't copy my mistakes. The build should be warning-free. :)I didn't notice any warnings using gcc 5.4.0.
I'm on 7.2.0, so that's probably why. It was "just" a style warning anyway. :)
Cheers,
Ulf
2018-02-21 5:09 GMT+09:00 Ulf Magnusson <[email protected]>:
Quoted 49 lines · Eugeniu Rosca
Quoted 27 lines · Ulf Magnusson
Quoted 26 lines · Masahiro Yamada
Quoted 8 lines
2018-02-20 17:13 GMT+09:00 Masahiro Yamada <[email protected]>:
Quoted 6 lines
I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.Please rewrite the code within 20 lines.
Quoted 8 lines
For a hint, I cleaned up the code base.
https://patchwork.kernel.org/patch/10229545/which should be equivalent to yours:
https://patchwork.kernel.org/patch/10226951/No 'enum
print_type', please.The reason I prefer them on separate lines consistently is to avoid stuff like the following:
Selected by [y]:
MV_XOR_V2[=y] && DMADEVICES [=y] && ARM64 [=y] Selected by [n]:
-AMCC_PPC440SPE_ADMA[=n] && DMADEVICES [=y] && (440SPe || ...
-FSL_RAID[=n] && DMADEVICES [=y] &&FSL_SOC&& ...
-INTEL_IOATDMA[=n] && DMADEVICES [=y] && PCI [=y] &&X86_64Quoted 5 lines
Quoted 1 line
That looks confusing and unbalanced to me.
I think nobody is disputing this. Masahiro's
https://patchwork.kernel.org/patch/10229545/ is also showing every reverse dependency top level OR expression on a new row.Ohh... sorry, didn't read closely enough.
Yeah, I agree that that looks like a nice and neat solution.
Quoted 11 lines
Quoted 10 lines
There are some simple ways to trim down the size of this patchset though.
Eugeniu:
What do you think about the following refactoring of your 3/3 patch? Maybe there's a way to haveexpr_print_revdep()take just a tristate too, though it's not worth it if it just grows the code elsewhere.
Quoted 5 lines
Quoted 2 lines
Well, Masahiro requests not using 'enum
print_type' and I still see it in your example.One alternative would be to keep the 'revdep' flag and add an extra tristate parameter for the kinds of selects to print.
Using 'enum tristate' is what was in mind, but I do not like to mess up __expr_print() any more.
I re-implemented this.
https://patchwork.kernel.org/patch/10231295/
Quoted 24 lines
I'm not a huge fan of having parameters that turn meaningless depending on the values of other parameters, but it might not be terrible there, if we must get rid of 'enum
print_type'. I'm always open to other suggestions too.The whole expression printing code feels a bit twisty to me with all the mutual recursion and stuff going on, but that's outside the scope of this patchset. :)
Quoted 6 lines
Quoted 4 lines
(By the way, I noticed that
expr_print_revdep()previously generated a warning suggesting parentheses, which was my fault. If you saw that, don't copy my mistakes. The build should be warning-free. :)I didn't notice any warnings using gcc 5.4.0.
I'm on 7.2.0, so that's probably why. It was "just" a style warning anyway. :)
Cheers,
Ulf
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Signature · 2 lines
Best Regards Masahiro Yamada
Quoted 53 lines · Masahiro Yamada
2018-02-21 5:09 GMT+09:00 Ulf Magnusson <[email protected]>:
Quoted 52 lines · Eugeniu Rosca
Quoted 27 lines · Ulf Magnusson
Quoted 26 lines · Masahiro Yamada
Quoted 8 lines
2018-02-20 17:13 GMT+09:00 Masahiro Yamada <[email protected]>:
Quoted 6 lines
I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.Please rewrite the code within 20 lines.
Quoted 8 lines
For a hint, I cleaned up the code base.
https://patchwork.kernel.org/patch/10229545/which should be equivalent to yours:
https://patchwork.kernel.org/patch/10226951/No 'enum
print_type', please.The reason I prefer them on separate lines consistently is to avoid stuff like the following:
Selected by [y]:
MV_XOR_V2[=y] && DMADEVICES [=y] && ARM64 [=y] Selected by [n]:
-AMCC_PPC440SPE_ADMA[=n] && DMADEVICES [=y] && (440SPe || ...
-FSL_RAID[=n] && DMADEVICES [=y] &&FSL_SOC&& ...
-INTEL_IOATDMA[=n] && DMADEVICES [=y] && PCI [=y] &&X86_64Quoted 5 lines
Quoted 1 line
That looks confusing and unbalanced to me.
I think nobody is disputing this. Masahiro's
https://patchwork.kernel.org/patch/10229545/ is also showing every reverse dependency top level OR expression on a new row.Ohh... sorry, didn't read closely enough.
Yeah, I agree that that looks like a nice and neat solution.
Quoted 13 lines
Quoted 10 lines
There are some simple ways to trim down the size of this patchset though.
Eugeniu:
What do you think about the following refactoring of your 3/3 patch? Maybe there's a way to haveexpr_print_revdep()take just a tristate too, though it's not worth it if it just grows the code elsewhere.Well, Masahiro requests not using 'enum
print_type' and I still see it in your example.
Quoted 9 lines
Quoted 2 lines
One alternative would be to keep the 'revdep' flag and add an extra tristate parameter for the kinds of selects to print.
Using 'enum tristate' is what was in mind, but I do not like to mess up
__expr_print()any more.I re-implemented this.
https://patchwork.kernel.org/patch/10231295/
I've tested https://patchwork.kernel.org/patch/10229545/ and
https://patchwork.kernel.org/patch/10231295/ and they work great for me. Thank you for this feature.
Quoted 19 lines
Quoted 19 lines
I'm not a huge fan of having parameters that turn meaningless depending on the values of other parameters, but it might not be terrible there, if we must get rid of 'enum
print_type'. I'm always open to other suggestions too.The whole expression printing code feels a bit twisty to me with all the mutual recursion and stuff going on, but that's outside the scope of this patchset. :)
Quoted 6 lines
Quoted 4 lines
(By the way, I noticed that
expr_print_revdep()previously generated a warning suggesting parentheses, which was my fault. If you saw that, don't copy my mistakes. The build should be warning-free. :)I didn't notice any warnings using gcc 5.4.0.
Quoted 14 lines
Quoted 8 lines
I'm on 7.2.0, so that's probably why. It was "just" a style warning anyway. :)
Cheers,
Ulf
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.htmlSignature · 2 lines
Best Regards Masahiro Yamada
Best regards,
Eugeniu.
Eugeniu,
2018-02-21 15:10 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 5 lines
I've tested https://patchwork.kernel.org/patch/10229545/ and
https://patchwork.kernel.org/patch/10231295/ and they work great for me. Thank you for this feature.
I will give you that code in the second one.
Please fill the commit log with yours,
and also fix the bugs pointed by Petr.
Anyway, I will claim my contribution
in the form of Signed-off-by when I pick it up. :)
Signature · 2 lines
Best Regards Masahiro Yamada
Hi Masahiro,
Quoted 6 lines · Masahiro Yamada
Eugeniu,
2018-02-21 15:10 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 1 line
Quoted 6 lines
Quoted 4 lines
I've tested https://patchwork.kernel.org/patch/10229545/ and
https://patchwork.kernel.org/patch/10231295/ and they work great for me. Thank you for this feature.I will give you that code in the second one.
I appreciate that, although I am open minded who takes the authorship. This feature makes my life easier and that's the most important part.
Quoted 3 lines
Please fill the commit log with yours,
and also fix the bugs pointed by Petr.
Sure. Will do that.
Quoted 2 lines
Anyway, I will claim my contribution
in the form of Signed-off-by when I pick it up. :)
I have one side question loosely related to this patch.
Reverse dependencies are still not printed in zconfdump() (used in our kernel team for e.g. cross platform .config alignment/comparison). My gut feeling tells me this is mainly caused by the extremely unfriendly unreadable expressions which is how reverse dependencies used to be represented before commit 1ccb27143360 ("kconfig: make "Selected by:" and "Implied by:" readable"). Since this is now greatly improved, would you mind printing Selected/Implied-by in zconfdump? There are also other zconfdump fixes sitting in my local branch, like https://patchwork.kernel.org/patch/9253535/ . I would like to push them to you (they probably deserve a separate thread), if you think this is appropriate.
Best regards,
Eugeniu.
Quoted 42 lines · Eugeniu Rosca
Hi Masahiro,
Quoted 6 lines · Masahiro Yamada
Eugeniu,
2018-02-21 15:10 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 1 line
Quoted 6 lines
Quoted 4 lines
I've tested https://patchwork.kernel.org/patch/10229545/ and
https://patchwork.kernel.org/patch/10231295/ and they work great for me. Thank you for this feature.I will give you that code in the second one.
I appreciate that, although I am open minded who takes the authorship. This feature makes my life easier and that's the most important part.
Quoted 3 lines
Please fill the commit log with yours,
and also fix the bugs pointed by Petr.Sure. Will do that.
Quoted 2 lines
Anyway, I will claim my contribution
in the form of Signed-off-by when I pick it up. :)I have one side question loosely related to this patch.
Reverse dependencies are still not printed in
zconfdump()(used in our kernel team for e.g. cross platform .config alignment/comparison). My gut feeling tells me this is mainly caused by the extremely unfriendly unreadable expressions which is how reverse dependencies used to be represented before commit1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable"). Since this is now greatly improved, would you mind printing Selected/Implied-by in zconfdump? There are also other zconfdump fixes sitting in my local branch, like https://patchwork.kernel.org/patch/9253535/ . I would like to push them to you (they probably deserve a separate thread), if you think this is appropriate.
Quoted 2 lines
Best regards,
Eugeniu.
Shameless plug:
If you need to do any fancy Kconfig parsing and comparison, then Kconfiglib (https://github.com/ulfalizer/Kconfiglib) might be handy. It makes it easy to walk the menu tree, print and inspect symbols and expressions, etc.
Cheers,
Ulf
Hi Ulf,
Quoted 46 lines · Ulf Magnusson
Quoted 42 lines · Eugeniu Rosca
Hi Masahiro,
Quoted 6 lines · Masahiro Yamada
Eugeniu,
2018-02-21 15:10 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 1 line
Quoted 6 lines
Quoted 4 lines
I've tested https://patchwork.kernel.org/patch/10229545/ and
https://patchwork.kernel.org/patch/10231295/ and they work great for me. Thank you for this feature.I will give you that code in the second one.
I appreciate that, although I am open minded who takes the authorship. This feature makes my life easier and that's the most important part.
Quoted 3 lines
Please fill the commit log with yours,
and also fix the bugs pointed by Petr.Sure. Will do that.
Quoted 2 lines
Anyway, I will claim my contribution
in the form of Signed-off-by when I pick it up. :)I have one side question loosely related to this patch.
Reverse dependencies are still not printed in
zconfdump()(used in our kernel team for e.g. cross platform .config alignment/comparison). My gut feeling tells me this is mainly caused by the extremely unfriendly unreadable expressions which is how reverse dependencies used to be represented before commit1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable"). Since this is now greatly improved, would you mind printing Selected/Implied-by in zconfdump? There are also other zconfdump fixes sitting in my local branch, like https://patchwork.kernel.org/patch/9253535/ . I would like to push them to you (they probably deserve a separate thread), if you think this is appropriate.Quoted 2 lines
Best regards,
Eugeniu.
Quoted 6 lines
Shameless plug:
If you need to do any fancy
Kconfigparsing and comparison, then Kconfiglib (https://github.com/ulfalizer/Kconfiglib) might be handy. It makes it easy to walk the menu tree, print and inspect symbols and expressions, etc.
Currently, when comparing kernel configurations of two platforms P1 and P2, our zconfdump parser generates reports like below. It helps us quite much during .config alignment. The report can be isolated to a specific menu entry (in this case, it is "General setup"). In this particular case, P1 is imx6 using v3.14 kernel and P2 is rcar3 using a v4.x kernel. If Kconfiglib is able to generate a similar diff report, I will happily embrace it. If not, we might have no choice but keep using zconfdump for a while.
Unfortunately, I haven't found some time to put my hands on Kconfiglib yet and I don't know what it is (or would be) capable of. Bottomline, the motivation to fix and enrich vanilla zconfdump functionality originates from making such reports possible on our side.
Boolean - - String
Tristate - \ / - Hex
config - \ || / - Integer
menuconfig - \ |||| /
\| |||||
Config Name Val MC TBSHI Cfg.Origin Comments
P1 P2 P1 P2 P1 P2 P1 P2 P1 P2
HAVE_KERNEL_GZIP y - C C B B sel-by - 01 -
HAVE_KERNEL_LZMA y - C C B B sel-by - 02 -
HAVE_KERNEL_XZ y - C C B B sel-by - 03 -
HAVE_KERNEL_LZO y - C C B B sel-by - 04 -
HAVE_KERNEL_LZ4 y - C C B B sel-by - 05 -
HAVE_KERNEL_LZ77 y - C - B - sel-by - 06 -
KERNEL_LZO y - C C B B defconfig - - -
KTIME_SCALAR y - C - B - sel-by - 07 -
GENERIC_CLOCKEVENTS_BUILD y - C - B - default - 08 -
IKCONFIG m y C C T T defconfig defconfig - -
MM_OWNER y - C - B - sel-by - 09 -
PERF_USE_VMALLOC y - C C B B sel-by - 10 -
HAVE_OPROFILE y - C C B B sel-by - 11 -
UPROBES y n C C B B sel-by default 12 13
ARCH_USE_BUILTIN_BSWAP y - C C B B sel-by - 14 -
HAVE_DMA_ATTRS y - C - B - sel-by - 15 -
ARCH_WANT_IPC_PARSE_VERSION y - C C B B sel-by - 16 -
MODULES_USE_ELF_REL y - C C B B sel-by - 17 -
OLD_SIGACTION y - C C B B sel-by - 18 -
TREE_PREEMPT_RCU y - C - B - sel-by-ch - 19 -
RCU_FANOUT 32 64 C C I I default default 20 21
RESOURCE_COUNTERS y - C - B - defconfig - - -Comments:
01) selected by (ARM [=y])
02) selected by (ARM [=y])
03) selected by (ARM [=y])
04) selected by (ARM [=y])
05) selected by (ARM [=y])
06) selected by (ARM [=y])
07) selected by (ARM [=y])
08) default y if GENERIC_CLOCKEVENTS [=y]
09) selected by (MEMCG [=y] && CGROUPS [=y] && RESOURCE_COUNTERS [=y])
10) selected by (ARM [=y])
11) selected by (ARM [=y] && HAVE_PERF_EVENTS [=y])
12) selected by (UPROBE_EVENT [=y] && TRACING_SUPPORT [=y] && FTRACE [=y] &&
ARCH_SUPPORTS_UPROBES [=y] && MMU [=y] && PERF_EVENTS [=y])
13) default n
14) selected by (ARM [=y])
15) selected by (ARM [=y])
16) selected by (ARM [=y])
17) selected by (ARM [=y])
18) selected by (ARM [=y])
19) selected by choice "RCU Implementation"
20) default 32 if ((TREE_RCU || TREE_PREEMPT_RCU) && !64BIT) [=y] 21) default 64 if ((TREE_RCU || PREEMPT_RCU) && RCU_EXPERT && 64BIT) [=y]
Quoted 2 lines
Cheers,
Ulf
Best regards,
Eugeniu.
Quoted 120 lines · Eugeniu Rosca
Hi Ulf,
Quoted 46 lines · Ulf Magnusson
Quoted 42 lines · Eugeniu Rosca
Hi Masahiro,
Quoted 6 lines · Masahiro Yamada
Eugeniu,
2018-02-21 15:10 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 1 line
Quoted 6 lines
Quoted 4 lines
I've tested https://patchwork.kernel.org/patch/10229545/ and
https://patchwork.kernel.org/patch/10231295/ and they work great for me. Thank you for this feature.I will give you that code in the second one.
I appreciate that, although I am open minded who takes the authorship. This feature makes my life easier and that's the most important part.
Quoted 3 lines
Please fill the commit log with yours,
and also fix the bugs pointed by Petr.Sure. Will do that.
Quoted 2 lines
Anyway, I will claim my contribution
in the form of Signed-off-by when I pick it up. :)I have one side question loosely related to this patch.
Reverse dependencies are still not printed in
zconfdump()(used in our kernel team for e.g. cross platform .config alignment/comparison). My gut feeling tells me this is mainly caused by the extremely unfriendly unreadable expressions which is how reverse dependencies used to be represented before commit1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable"). Since this is now greatly improved, would you mind printing Selected/Implied-by in zconfdump? There are also other zconfdump fixes sitting in my local branch, like https://patchwork.kernel.org/patch/9253535/ . I would like to push them to you (they probably deserve a separate thread), if you think this is appropriate.Quoted 2 lines
Best regards,
Eugeniu.Quoted 6 lines
Shameless plug:
If you need to do any fancy
Kconfigparsing and comparison, then Kconfiglib (https://github.com/ulfalizer/Kconfiglib) might be handy. It makes it easy to walk the menu tree, print and inspect symbols and expressions, etc.Currently, when comparing kernel configurations of two platforms P1 and P2, our zconfdump parser generates reports like below. It helps us quite much during .config alignment. The report can be isolated to a specific menu entry (in this case, it is "General setup"). In this particular case, P1 is imx6 using v3.14 kernel and P2 is rcar3 using a v4.x kernel. If Kconfiglib is able to generate a similar diff report, I will happily embrace it. If not, we might have no choice but keep using zconfdump for a while.
Unfortunately, I haven't found some time to put my hands on Kconfiglib yet and I don't know what it is (or would be) capable of. Bottomline, the motivation to fix and enrich vanilla zconfdump functionality originates from making such reports possible on our side.
Boolean - - String Tristate - \ / - Hex config - \ || / - Integer menuconfig - \ |||| / \| ||||| Config Name Val MC TBSHI Cfg.Origin Comments P1 P2 P1 P2 P1 P2 P1 P2 P1 P2 HAVE_KERNEL_GZIP y - C C B B sel-by - 01 - HAVE_KERNEL_LZMA y - C C B B sel-by - 02 - HAVE_KERNEL_XZ y - C C B B sel-by - 03 - HAVE_KERNEL_LZO y - C C B B sel-by - 04 - HAVE_KERNEL_LZ4 y - C C B B sel-by - 05 - HAVE_KERNEL_LZ77 y - C - B - sel-by - 06 - KERNEL_LZO y - C C B B defconfig - - - KTIME_SCALAR y - C - B - sel-by - 07 - GENERIC_CLOCKEVENTS_BUILD y - C - B - default - 08 - IKCONFIG m y C C T T defconfig defconfig - - MM_OWNER y - C - B - sel-by - 09 - PERF_USE_VMALLOC y - C C B B sel-by - 10 - HAVE_OPROFILE y - C C B B sel-by - 11 - UPROBES y n C C B B sel-by default 12 13 ARCH_USE_BUILTIN_BSWAP y - C C B B sel-by - 14 - HAVE_DMA_ATTRS y - C - B - sel-by - 15 - ARCH_WANT_IPC_PARSE_VERSION y - C C B B sel-by - 16 - MODULES_USE_ELF_REL y - C C B B sel-by - 17 - OLD_SIGACTION y - C C B B sel-by - 18 - TREE_PREEMPT_RCU y - C - B - sel-by-ch - 19 - RCU_FANOUT 32 64 C C I I default default 20 21 RESOURCE_COUNTERS y - C - B - defconfig - - -Comments:
01) selected by (ARM [=y])
02) selected by (ARM [=y])
03) selected by (ARM [=y])
04) selected by (ARM [=y])
05) selected by (ARM [=y])
06) selected by (ARM [=y])
07) selected by (ARM [=y])
08) default y ifGENERIC_CLOCKEVENTS[=y]
09) selected by (MEMCG [=y] && CGROUPS [=y] &&RESOURCE_COUNTERS[=y])
10) selected by (ARM [=y])
11) selected by (ARM [=y] &&HAVE_PERF_EVENTS[=y])
12) selected by (UPROBE_EVENT[=y] &&TRACING_SUPPORT[=y] && FTRACE [=y] &&ARCH_SUPPORTS_UPROBES [=y] && MMU [=y] && PERF_EVENTS [=y])13) default n
14) selected by (ARM [=y])
15) selected by (ARM [=y])
16) selected by (ARM [=y])
17) selected by (ARM [=y])
18) selected by (ARM [=y])
19) selected by choice "RCU Implementation"
Quoted 2 lines
20) default 32 if ((TREE_RCU || TREE_PREEMPT_RCU) && !64BIT) [=y] 21) default 64 if ((TREE_RCU || PREEMPT_RCU) && RCU_EXPERT && 64BIT) [=y]
Think it would be a good fit for that. There's a load_config() function for loading a .config, and after that you can inspect symbol values, properties, expressions, etc.
The menu tree is exposed directly too, so you could look for a menu with a particular name and limit yourself to that. See e.g.
https://github.com/ulfalizer/Kconfiglib/blob/master/examples/print_tree.py
for how to walk the menu tree.
Kconfiglib is kinda like an "open-internals" version of the C tools if you will, and pretty flexible.
I'm starting to feel a bit dirty, so I guess we could move it to private mail or an issue on the GitHub page or something if you give it a try and have any questions. Think it would work well at least. :)
Cheers,
Ulf
Quoted 70 lines · Masahiro Yamada
2018-02-20 17:13 GMT+09:00 Masahiro Yamada <[email protected]>:
Quoted 69 lines
2018-02-19 5:47 GMT+09:00 Eugeniu Rosca <[email protected]>:
Quoted 57 lines
From: Eugeniu Rosca <[email protected]>
Hello Masahiro, Ulf, Petr and all,
Here are a few words about the motivation behind this patch series.
First, the reason I got in touch with
Kconfigis to optimize the configuration of automotive kernels, as well as to align the kernel configuration across a number of platforms. In the context of kernel optimization, one of the primary goals is to filter out any features that are not mentioned in the platform requirements.Surprisingly or not, disabling a CONFIG option (which is assumed to be unneeded) may be not so trivial. Especially it is not trivial, when this CONFIG option is selected by a dozen of other configs. Before the moment commit
1ccb27143360("kconfig: make "Selected by:" and "Implied by:" readable") was submitted by Petr and eventually popped up in v4.16-rc1, it was an absolute pain to break down the "Selected by" reverse dependency expression in order to identify all those configs which select (IOW *do not allow disabling*) a certain feature (assumed to be not needed).This patch series tries to make one step further and puts at users' fingertips the revdep top level OR sub-expressions grouped/clustered by the tristate value they evaluate to. This should allow the users to directly concentrate on and tackle the active reverse dependencies, which imho are the only ones that matter for a given ARCH and for a given defconfig (nevertheless we still print all of them).
Changes
v3->v4(fixed review findings from Ulf):
- Remove redundant default cases in switch constructs.
- Remove gettext_()tokens instr_append()calls.
- Aggregate code repetitions inexpr_print_revdep().Changes
v2->v3:
- Switch from reverse dependencies prefixed by their tristate value to reverse dependencies grouped by the tristate value they evaluate to.
- Skip printing "{Selected,Implied} by [y|m|n]:" if there are no top level OR tokens/sub-expressions that evaluate to y|m|n (suggested by Petr).
- Use [1] as template for updating the interface/prototype of__expr_print()(suggested by Ulf).Changes
v1->v2:
- Don't skip the =n reverse dependency OR tokens, since some users might still need this information (suggested by Ulf).
- Instead of using "Selected by" for active tokens only, use it for all OR tokens, but specify the tristate value of each token as prefix (suggested by Masahiro).[1] https://marc.info/?l=linux-kbuild&m=151777006005199&w=4
Eugeniu Rosca (3):
kconfig: Print reverse dependencies on new line consistently
kconfig: Prepare for printing reverse dependencies in groups
kconfig: Print reverse dependencies in groupsQuoted 6 lines
scripts/kconfig/expr.c | 102 +++++++++++++++++++++++++++++++++++++------- scripts/kconfig/expr.h | 11 ++++- scripts/kconfig/lkc_proto.h | 1 + scripts/kconfig/menu.c | 37 +++++++++++----- 4 files changed, 124 insertions(+), 27 deletions(-)I do not like this implementation.
The code is super ugly, the diff-stat is too much than needed.
Quoted 8 lines
Quoted 2 lines
Please rewrite the code within 20 lines.
For a hint, I cleaned up the code base.
https://patchwork.kernel.org/patch/10229545/which should be equivalent to yours:
https://patchwork.kernel.org/patch/10226951/
Fwiw CP reports for https://patchwork.kernel.org/patch/10229545/
WARNING: unnecessary whitespace before a quoted newline #113: FILE: scripts/kconfig/menu.c:830:
+ str_append(r, _(" Selected by: \n"));
WARNING: unnecessary whitespace before a quoted newline #121: FILE: scripts/kconfig/menu.c:836:
+ str_append(r, _(" Implied by: \n"));
Both solutions behave the same for me.
I agree Masahiro's patch is much nicer.
Fwiw, both solutions display reverse dependencies of certain choices like below:
Prompt: Compiler optimization level
Location:
-> General setup
Defined at init/Kconfig:1021
Selected by:
- m
I can notice this output at least for:
prompt "Cputime accounting"
prompt "Compiler optimization level"
prompt "Choose SLAB allocator"
prompt "Processor family"
prompt "Preemption Model"
prompt "Timer frequency"
prompt "IO delay type"
prompt "Choose kernel unwinder"
prompt "Default security module"
I'm not sure if Selected by:\n - m is of any use/help. Omitting it, if desired, is probably subject of a different patch?
Quoted 1 line
No 'enum
print_type', please.
Ok. Feedback is appreciated.
Quoted 5 lines
Signature · 3 lines
-- Best Regards Masahiro Yamada
Best regards,
Eugeniu.
Quoted 21 lines · Eugeniu Rosca
Fwiw, both solutions display reverse dependencies of certain choices like below:
Prompt: Compiler optimization level
Location:-> General setupDefined at init/Kconfig:1021
Selected by:
- mI can notice this output at least for:
prompt "Cputime accounting"
prompt "Compiler optimization level"
prompt "Choose SLAB allocator"
prompt "Processor family"
prompt "Preemption Model"
prompt "Timer frequency"
prompt "IO delay type"
prompt "Choose kernel unwinder"
prompt "Default security module"
Quoted 2 lines
I'm not sure if
Selected by:\n - mis of any use/help. Omitting it, if desired, is probably subject of a different patch?
This is due to some obscure Kconfig internals. I added a comment in scripts/kconfig/menu.c to explain it recently:
/*
* For non-optional choices, add a reverse dependency (corresponding to
* a select) of '<visibility> && m'. This prevents the user from
* setting the choice mode to 'n' when the choice is visible.
*
* This would also work for non-choice symbols, but only non-optional
* choices clear SYMBOL_OPTIONAL as of writing. Choices are implemented
* as a type of symbol.
*/
if (sym && !sym_is_optional(sym) && parent->prompt) {
sym->rev_dep.expr = expr_alloc_or(sym->rev_dep.expr,
expr_alloc_and(parent->prompt->visible.expr,
expr_alloc_symbol(&symbol_mod)));
}It doesn't hurt to show it, IMO. Might cause some confusion at worst, but it's probably pretty minimal.
Cheers,
Ulf